An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase

Katja Venko; Marjana Novič

doi:10.3390/molecules24050837

An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase

Venko, Katja;Novič, Marjana 2019-02-27 00:00:00 molecules Article An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase Katja Venko * and Marjana Novic ˇ * Laboratory for Cheminformatics, Theory Department, National Institute of Chemistry, SI-1000 Ljubljana, Slovenia * Correspondence: [email protected] (K.V.); [email protected] (M.N.); Tel.: +386-1-4760253 (M.N.) Received: 1 February 2019; Accepted: 26 February 2019; Published: 27 February 2019 Abstract: Phenols are the most abundant naturally accessible antioxidants present in a human normal diet. Since numerous beneﬁcial applications of phenols as preventive agents in various diseases were revealed, the evaluation of phenols bioavailability is of high interest of researchers, consumers and drug manufacturers. The hydrophilic nature of phenols makes a cell membrane penetration difﬁcult, which imply an alternative way of uptake via membrane transporters. However, the structural and functional data of membrane transporters are limited, thus the in silico modelling is really challenging and urgent tool in elucidation of transporter ligands. Focus of this research was a particular transporter bilitranslocase (BTL). BTL has a broad tissue expression (vascular endothelium, absorptive and excretory epithelia) and can transport wide variety of poly-aromatic compounds. With available BTL data (pKi [mmol/L] for 120 organic compounds) a robust and reliable QSAR models for BTL transport activity were developed and extrapolated on 300 phenolic compounds. For all compounds the transporter proﬁles were assessed and results show that dietary phenols and some drug candidates are likely to interact with BTL. Moreover, synopsis of predictions from BTL models and hits/predictions of 20 transporters from Metrabase and Chembench platforms were revealed. With such joint transporter analyses a new insights for elucidation of BTL functional role were acquired. Regarding limitation of models for virtual proﬁling of transporter interactions the computational approach reported in this study could be applied for further development of reliable in silico models for any transporter, if in vitro experimental data are available. Keywords: phenols; membrane transporters; bilitranslocase; transport activity; QSAR; in silico models 1. Introduction Phenols are one of the most promising groups of dietary preventive agents [1,2]. They are widely amenable natural and nutritional agents, i.e. nutraceuticals, which belong to a heterogeneous family of chemical compounds comprising benzoquinones, phenolic acids and aldehydes, coumarins, chromones, xanthonoids, stilbenoids, ﬂavonoids, lignans, tannins, etc. [3]. In general, phenolic compounds are classiﬁed based on the number of phenol units in the molecule, as simple phenols or polyphenols [2,4]. Functionally, they are secondary metabolites produced by plants and microorganisms or synthesized industrially [5,6]. As the most abundant antioxidants present in a human normal diet they have received increasing interest from researchers, consumers, food and drug manufacturers [2,7]. The phenols, being recognized as one of the most potential antioxidants, have role in preventing an imbalance in reactive oxygen species (ROS) production and an attenuation Molecules 2019, 24, 837; doi:10.3390/molecules24050837 www.mdpi.com/journal/molecules Molecules 2019, 24, 837 2 of 21 of anti-oxidant defense, what can turnout the genesis of many ageing phenomena and diseases [1,8–10]. Furthermore, bioactivities of phenols appeared to be quite broad-spanning, yet many multi-step interactions have not been exactly elucidated. Thus, numerous beneﬁcial application of phenols as preventive and therapeutic agents in cancer and inﬂammatory, neurodegenerative and cardiovascular diseases have been reported [1,9,11–32]. Particularly, the initiation stages of carcinogenesis were shown to be largely preventable and susceptible to modulation by nutraceuticals, especially highlighting the dietary polyphenols as one of the most promising anti-carcinogenic agents [10,15]. However, solely the normal diet phenolic intake is insufﬁcient for the effective chemoprevention, but the ingestion of extra dietary phenolic supplements as support to chemotherapy is a promising approach [7,10]. The epidemiological data show that life-long intake of phenolic compounds is beneﬁcial and related to reduce the incidence of various diseases [8,11,15,33], yet to improve our knowledge on the bioavailability of polyphenols from the diet uptake is of high interest. The uptake of nutraceuticals and orally administrated drugs is dependent on their capacity of absorption from the digestive tract. The studies, which followed the oral ingestion and gastro-intestinal uptake of ﬂavonoids, show that the concentrations are <1% of the administrated dose [34,35]. Such stringent intestinal control can be explained with the experimental evidence that many importers are unable to transport phenols, while various exporters can [34]. This complexity obscures the delivery of phenolic nutraceuticals or drugs, and raises the need for good substrate permeability predictive models. Since phenols are widely distributed in fruits, vegetables, grains and their processed foodstuffs, but are not synthesized by animal cells, their detection in animals is solely indicative of plant ingestion [36]. Therefore, their absorption and tissue distribution in specimens is indispensably dependent on transport ability of membrane transporters, and in this regard the cellular uptake of phenolic compounds can be efﬁciently estimated by in vitro transport assays [37–39], while in silico applications of membrane transport processes are very limited due to extremely slow rise of high-resolution structures of membrane transporters [40]. At present, the development of in silico methodologies for ﬁlling the gap among protein structures and properties is emerging and helps circumventing the difﬁculties in resolving membrane transporter structures and functions [40–42]. In general, two in silico strategies exists: (i) ligand-based methods enabling analysis of transporter substrate speciﬁcity in the absence of 3D structure, and (ii) protein-based methods providing modeling of transporters structures based on sequence identity of available experimentally determined atomic structures [43]. Among ﬁrst, the pharmacophore mapping and development of quantitative structure-activity relationship (QSAR) models are the most common for predicting ligand binding modes (Figure 1) [44]. Among second, the comparative modeling is the most frequently used approach to reveal structure and function of transporter, yet is only applicable, if homolog structures are available [45]. Furthermore, studies show that the combination of both strategies is favorable as it provides information beyond individual models [45]. Due to shortage of currently available experimental functional and structural data on membrane transporters, the in silico modelling is essential in the discovery of adverse effects, ADMET properties and more efﬁcient transporter substrates (e.g., dietary products, drugs) [44]. Molecules 2019, 24, x FOR PEER REVIEW 3 of 21 transporters studies. Nevertheless, according to the reports in TransportDB (www.membranetransport.org) [47] and Transporter Classification Database (www.tcdb.org) [48] a hundreds of potential transporters exist (in the human genome > 1400 unique transport proteins were identified). Therefore, our interest arises on a specific, neglected but extremely attractive Molecules 2019, 24, 837 3 of 21 transporter bilitranslocase (BTL, Uniprot O88750, GenBank: Y12178.1). Figure Figure 1. 1. General scheme o General scheme of f the quantitative structur the quantitative structure-activity relatio e-activity relationsh nship (QSAR) ip (QSAR) strategy. strategy. Chemical structures and biological activity data are obligated input for model building and Chemical structures and biological activity data are obligated input for model building and validation. After validat the ion. model After the mode development l de and velopment an optimization d optimizat the new ion activity the new act predictions ivitycan predicti be made ons c based an be made solely on chemical structure input. based solely on chemical structure input. By evidence dietary substances and drugs in mammals are transported via membrane Bilitranslocase is a 38.22 kDa bilirubin membrane transporter and is the first reported member transporters rather than passive diffusion [38]. Therefore, it is of interest to proﬁle the membrane of the Bilirubin Transporter (BRT) Family (TCDB 2.A.65.1.1) [35]. The gene of BTL is unidentified, transport ability by using the recently available transporter databases in the public domain like and even no homology with any other known protein structure was found [35,49]. Yet, a surprising Metrabase (www-metrabase.ch.cam.ac.uk) [46] and ChemBench (www.chembench.mml.unc.edu) [44]. feature of BTL is its coding sequence, which is 94% homologous to a segment of the antisense strand Metrabase provides the structural, physiochemical and biological data about transporters and their of the ceruloplasmin transcript (RNO 30106) in the anti-CODE database (www.anticode.org). The ligands. It contains 16 efﬂux and inﬂux transporters, which are also included in the guidance for considerable biological importance in BTL sequence has the bilirubin-binding motif, which plays a the approval of drugs proposed by the US Food and Drug Administration and by the European central role in ligand binding, and it is similar to highly conserved segment in α-phycocyanins [49], Medicines Agency [46]. On the other hand, Chembench platform offers the largest collection of which are ancient biliproteins present in cyanobacteria. Further on, BTL shares similar tissue human intestinal transporters interactions and the compendium of QSAR models for virtual screening distribution and functional properties with some of the members of the family of organic anion of chemicals. Moreover, these two databases reﬂect the current research focus in mammalian transporters (OAT) (Figure 2), which is the most dominant family of the solute carriers (SLC, one of transporters, as up to the best of our knowledge they encompass majority of existing transporters the two classified categories of transporters) [50]. BTL does not transport only bilirubin into studies. Nevertheless, according to the reports in TransportDB (www.membranetransport.org) [47] hepatocytes [51,52], but can transport a wide variety of poly-aromatic compounds among vitally and Transporter Classiﬁcation Database (www.tcdb.org) [48] a hundreds of potential transporters exist important endogenous substrates and drugs [35,53,54]. Its tissue expression is broad from vascular (in the human genome > 1400 unique transport proteins were identiﬁed). Therefore, our interest arises endothelium to stomach, intestine, kidney and liver epithelia (Figure 2) [55–58]. These peculiarities on a speciﬁc, neglected but extremely attractive transporter bilitranslocase (BTL, Uniprot O88750, make BTL a hot spot for future investigation as potential important endogenous drug delivery GenBank: Y12178.1). system to the intracellular targets, especially for its ability of cancer drugs uptake [59] and also as Bilitranslocase is a 38.22 kDa bilirubin membrane transporter and is the ﬁrst reported member tumor biomarker [60]. The study on renal carcinoma cells figure out an outstanding fact, that BTL of the Bilirubin Transporter (BRT) Family (TCDB 2.A.65.1.1) [35]. The gene of BTL is unidentiﬁed, expression is severely down-regulated in the renal cortical tumors [61], therefore the BTL assessment and even no homology with any other known protein structure was found [35,49]. Yet, a surprising with specific BTL antibodies could serve as a novel biomarker of human kidney cancers [60]. In spite feature of BTL is its coding sequence, which is 94% homologous to a segment of the antisense of our high interest and several efforts, the synthesis of BTL was unsuccessful, thus its structural and strand of the ceruloplasmin transcript (RNO 30106) in the anti-CODE database (www.anticode.org). functional characterization remains unclear. However, up to date the data on tissue expression The considerable biological importance in BTL sequence has the bilirubin-binding motif, which plays based on the specific BTL monoclonal antibodies [35], transport activity assays [53,54] and a central role in ligand binding, and it is similar to highly conserved segment in -phycocyanins [49], which are ancient biliproteins present in cyanobacteria. Further on, BTL shares similar tissue distribution and functional properties with some of the members of the family of organic anion transporters (OAT) (Figure 2), which is the most dominant family of the solute carriers (SLC, Molecules 2019, 24, 837 4 of 21 one of the two classiﬁed categories of transporters) [50]. BTL does not transport only bilirubin into hepatocytes [51,52], but can transport a wide variety of poly-aromatic compounds among vitally important endogenous substrates and drugs [35,53,54]. Its tissue expression is broad from vascular endothelium to stomach, intestine, kidney and liver epithelia (Figure 2) [55–58]. These peculiarities make BTL a hot spot for future investigation as potential important endogenous drug delivery system to the intracellular targets, especially for its ability of cancer drugs uptake [59] and also as tumor biomarker [60]. The study on renal carcinoma cells ﬁgure out an outstanding fact, that BTL expression is severely down-regulated in the renal cortical tumors [61], therefore the BTL assessment with speciﬁc BTL antibodies could serve as a novel biomarker of human kidney cancers [60]. In spite of our high interest and several efforts, the synthesis of BTL was unsuccessful, thus its structural and functional characterization remains unclear. However, up to date the data on tissue Molecules 2019, 24, x FOR PEER REVIEW 4 of 21 expression based on the speciﬁc BTL monoclonal antibodies [35], transport activity assays [53,54] and computational elucidation of its 3D structure based on available NMR data (Figure 2) [62] are computational elucidation of its 3D structure based on available NMR data (Figure 2) [62] are available. available Since BTL . Sin gene ce BTL is not g deﬁned, ene is not the de BTL fincharacterization ed, the BTL char experimentally acterization exp still eri rm emains entally limited still re only mains on limited only on immunological approach by specific ant-peptide antibodies and in silico modeling immunological approach by speciﬁc ant-peptide antibodies and in silico modeling with identiﬁed wi synthetic th identif transmembrane ied synthetic tra segments. nsmembrane segments. Figure 2. (a) Structural model of membrane transporter bilitranslocase (top-scoring computational Figure 2. (a) Structural model of membrane transporter bilitranslocase (top-scoring computational arrangement of four transmembrane helices based on NMR spectroscopy data) [62]. No homologs arrangement of four transmembrane helices based on NMR spectroscopy data) [62]. No homologs were were defined for 1026 bp mRNA sequence (GenBank: Y12178.1), thus the comparative modeling deﬁned for 1026 bp mRNA sequence (GenBank: Y12178.1), thus the comparative modeling studies studies are not possible to perform. The further structure characterization of bilitranslocase is under are not possible to perform. The further structure characterization of bilitranslocase is under study. study. (b) Scheme of cell membrane transporters distribution in different tissues [44,50,58]. ABC (b) Scheme of cell membrane transporters distribution in different tissues [44,50,58]. ABC transporters transporters ar are indicated e indicate by black arr d by ows black arrows a and solute carriers nd solute (SLC) carriers (SLC) by w by white arrows; hite arrows; on one direction e direction transport transport (one (one way arrow), way arrow), bidirectional tran bidirectional transport (two sport (two way way arrow). arrow). Considering the current limitation and dearth of empirical data about the interesting BTL transporter, the usage of in silico approaches is more than welcome to elucidate the BTL functional properties. Moreover, they can even provide new guidelines for further design of the BTL experiments. Some QSAR models for prediction of the BTL transport ability were preliminary build on available data of biological assay of the BTL transport activity for the poly-aromatic organic compounds [53,54,59]. The in vitro experiments showed fast kinetics of plasma appearance of polyphenols and their concentrations were found in the range of half-saturation of BTL in the portal blood [63]. Thus, Passamonti et al. suggest that particularly at the gastric level the BTL-mediated mechanism could be the basis of the fast kinetics and ability of polyphenols to permeate the gastric mucosa [35,63]. Even more, it seems that the fast absorption of polyphenols in the upper digestive tract is a mindful steep, as polyphenols in the lower segments of digestive tract are exposed to intensive chemical and microbial degradation [1,34], which significantly affect their bioavailability. The aim of this study was to assess the membrane transporters activity of selected phenolic compounds by a comprehensive virtual profiling. Firstly, we developed new robust and reliable models for BTL transport activity that were used for screening of new untested dataset of 300 Molecules 2019, 24, 837 5 of 21 Considering the current limitation and dearth of empirical data about the interesting BTL transporter, the usage of in silico approaches is more than welcome to elucidate the BTL functional properties. Moreover, they can even provide new guidelines for further design of the BTL experiments. Some QSAR models for prediction of the BTL transport ability were preliminary build on available data of biological assay of the BTL transport activity for the poly-aromatic organic compounds [53,54,59]. The in vitro experiments showed fast kinetics of plasma appearance of polyphenols and their concentrations were found in the range of half-saturation of BTL in the portal blood [63]. Thus, Passamonti et al. suggest that particularly at the gastric level the BTL-mediated mechanism could be the basis of the fast kinetics and ability of polyphenols to permeate the gastric mucosa [35,63]. Even more, it seems that the fast absorption of polyphenols in the upper digestive tract is a mindful steep, as polyphenols in the lower segments of digestive tract are exposed to intensive chemical and microbial degradation [1,34], which signiﬁcantly affect their bioavailability. The aim of this study was to assess the membrane transporters activity of selected phenolic compounds by a comprehensive virtual proﬁling. Firstly, we developed new robust and reliable models for BTL transport activity that were used for screening of new untested dataset of 300 phenolic compounds (including potential drug candidates, nutraceuticals). With the developed QSAR models the new insights into transport interactions of structurally unresolved BTL transmembrane protein were gained and the phenolic compounds that could inhibit the BTL transport were identiﬁed. We also discussed the preferences of using the consensus of models to enhance predictivity of in silico models. Secondly, the unique transporter proﬁle of studied compounds was performed based on the results from a new BTL models developed in this study and from the results obtained from Metrabase [46] and Chembench platforms [44], a freely available tools for membrane transporters characterization. 2. Results 2.1. BTL Classiﬁcation Models The development of QSAR models is a complex process therefore general OECD principles to properly produce and validate models [64,65] have been taken into account. In order to build reliable models and also to detect possible model pathologies in the modelling strategy, a cross validation and external validation was included. Thus, BTL dataset (n = 120) was divided into two or three subsets in rate 75/25 or 60/25/15. The model NN-C had three subsets; training set (n = 70), test set (n = 31) and validation set (n = 19). Models NN-D and Q-D had two subsets; training set (n = 90) and validation set (n = 30). A dataset splitting conditions are precisely stated in study of Martinci ˇ c ˇ et al. [59]. Initial modeling datasets included 66 or 78 variables, Codessa and Dragon descriptors, respectively. The model NN-C was the best model available from study of Martinci ˇ c ˇ et al. [59] and was developed with non-reduced number of descriptors (66 Codessa MDs). In this study new Dragon molecular descriptors (MDs) were calculated and further model optimization with cross validation and genetic algorithm was used. The newly developed models (NN-D and Q-D) contain signiﬁcantly reduced set of MDs (from 78 to 18/11). The list of selected descriptors of NN-D and Q-D models is represented in Table S4 (Supplementary Material). The selected models have comparable quality parameters for training set, yet new CP-ANN model has signiﬁcantly improved performance of validation set (Tables 1 and 2). Regarding results of quantitative quality indicators and graphical quality parameter (ROC curve) the NN-D model shows the best training and validation performances (Figure 3). Predictions for compounds used in the models development and validation are presented in Table S1 (Supplementary Material). Molecules 2019, 24, x FOR PEER REVIEW 5 of 21 phenolic compounds (including potential drug candidates, nutraceuticals). With the developed QSAR models the new insights into transport interactions of structurally unresolved BTL transmembrane protein were gained and the phenolic compounds that could inhibit the BTL transport were identified. We also discussed the preferences of using the consensus of models to Molecules 2019, 24, 837 6 of 21 enhance predictivity of in silico models. Secondly, the unique transporter profile of studied compounds was performed based on the results from a new BTL models developed in this study Table 1. Statistic parameters of the best three single models and consensus classiﬁcation models. and from the results obtained from Metrabase [46] and Chembench platforms [44], a freely available tools for membrane transporters characterization. Single Models 2-Model Consensus 3-Model Consensus NN-C + NN-C + NN-C + 2. Results NN-C NN-D Q-D A B A + B NN-D Q-D Q-D TP 44 44 46 33 33 41 29 19 48 2.1. BTL Classification Models TN 61 65 58 47 49 55 45 17 62 FP 9 5 12 6 2 1 1 7 8 The development of QSAR models is a complex process therefore general OECD principles to FN 6 6 4 0 2 0 0 2 2 properly produce and validate models [64,65] have been taken into account. In order to build 120 120 120 86 86 97 75 45 120 reliable models and also to detect possible model pathologies in the modelling strategy, a cross PR 1 (0.98) 1 (0.92) 1 (0.97) 0.72 0.72 0.81 0.62 0.37 1 validation and external validation was included. Thus, BTL dataset (n = 120) was divided into two or SE 0.88 0.88 0.92 1.00 0.94 1.00 1.00 0.90 0.96 three subsets SP in ra 0.87 te 75/25 0.93 or 60/25/0.83 15. The mo 0.89 del NN-0.96 C had three 0.98 subsets; tra 0.98 ini 0.71 ng set ( 0.89 n = 70), test ACC 0.88 0.91 0.87 0.93 0.95 0.99 0.99 0.80 0.92 set (n = 31) and validation set (n = 19). Models NN-D and Q-D had two subsets; training set (n = 90) NPV 0.91 0.92 0.94 1.00 0.96 1.00 1.00 0.89 0.97 and validation set (n = 30). A dataset splitting conditions are precisely stated in study of Martinčič et PPV 0.83 0.90 0.79 0.85 0.94 0.98 0.97 0.73 0.86 al. [59]. MCC 0.75 0.81 0.74 0.87 0.90 0.98 0.97 0.62 0.84 Initial modeling datasets included 66 or 78 variables, Codessa and Dragon descriptors, A—complete agreement among predictions, B—maximal one disagreement among predictions. respectively. The model NN-C was the best model available from study of Martinčič et al. [59] and was developed with non-reduced number of descriptors (66 Codessa MDs). In this study new Table 2. Summary of classiﬁcation models results. Dragon molecular descriptors (MDs) were calculated and further model optimization with cross No. of Active validation and genetic algorithm was used. The newly develope Predict d Ratio models (NN-D an [%] d Q-D) contain No. of Compound Predictions Type ACC ACC Model ID. TR V Descriptors significantly reduced set of MDs (from 78 to 18/11). The list of selected descriptors of NN-D and Q-D BTL Dataset Phenols BTL Dataset Phenols models is represented in Table S4 (Supplementary Material). The selected models have comparable NN-C CP-ANN 66 0.93 0.74 100 (97) 100 (94) 54 (51) 75 (74) quality parameters for training set, yet new CP-ANN model has significantly improved performance NN-D CP-ANN 18 0.92 0.87 100 (92) 100 (69) 49 (46) 138 (100) Q-D MLR 11 0.88 0.77 100 (97) 100 (93) 58 (55) 72 (68) of validation set (Tables 1 and 2). Regarding results of quantitative quality indicators and graphical A + B 3 cons / 0.93 0.87 100 (99) 100 (100) 56 (55) 65 (65) quality parameter (ROC curve) the NN-D model shows the best training and validation A 3 cons / 1.00 0.92 62 (62) 36 (36) 30 (30) 15 (15) performances (Figure 3). Predictions for compounds used in the models development and validation NN-D + Q-D 2 cons / 1.00 0.96 81 (81) 50 (50) 42 (42) 31 (31) are presented in T ( ) values ablefor S1 compounds (Supplement in AD; ary M 3 cons—3-model aterial). consensus; 2 cons—2-model consensus. Figure 3. ROC curves of the three selected classiﬁcation models: (a) training set, (b) validation set. Figure 3. ROC curves of the three selected classification models: (a) training set, (b) validation set. The reliability of a QSAR model depends on its capacity to achieve conﬁdent predictions of new The reliability of a QSAR model depends on its capacity to achieve confident predictions of new compounds not considered in the model development, which implies the necessity of considering compounds not considered in the model development, which implies the necessity of considering the applicability domain (AD) of the model. Several AD analyses including the descriptor ’s range, the applicability domain (AD) of the model. Several AD analyses including the descriptor’s range, leverages, Euclidian distances, rivality and modelability indexes [66–68] are available. In this study we leverages, Euclidian distances, rivality and modelability indexes [66–68] are available. In this study used two of AD approaches; maximal Euclidian distance of training set in NN-C (EDcrt = 6.16) and we used two of AD approaches; maximal Euclidian distance of training set in NN-C (EDcrt = 6.16) NN-D model (ED = 2.36), and leverage values in Q-D model (hat* = 0.4) (Figure 4) [67,68]. crt and NN-D model (EDcrt = 2.36), and leverage values in Q-D model (hat* = 0.4) (Figure 4) [67,68]. Molecules 2019, 24, 837 7 of 21 Molecules 2019, 24, x FOR PEER REVIEW 6 of 21 Figure 4. Graphical assessment of the applicability domains of the best three classiﬁcation models Figure 4. Graphical assessment of the applicability domains of the best three classification models (NN-C, NN-D, Q-D). Compounds with estimated Euclidian distances/leverages > ED /hat* fail out (NN-C, NN-D, Q-D). Compounds with estimated Euclidian distances/leverages > EDcrit/hat* fail out crit of applicability domain of model. of applicability domain of model. The structure activity relationship analysis has revealed the inﬂuential structural properties of BTL The structure activity relationship analysis has revealed the influential structural properties of ligands. In general, high capability of compounds of forming hydrogen bonds, ionic properties and BTL ligands. In general, high capability of compounds of forming hydrogen bonds, ionic properties planarity of molecules are the most relevant one. The same features were highlighted in previous BTL and planarity of molecules are the most relevant one. The same features were highlighted in experimental studies [53,54,59]. The molecular descriptors, which are used in newly developed previous BTL experimental studies [53,54,59]. The molecular descriptors, which are used in newly models NN-D and Q-D (Table S4, Supplementary Material) represent the following properties: developed models NN-D and Q-D (Table S4, Supplementary Material) represent the following planarity (number of multiple and aromatic bonds — nBM, nArCO, nArOR, ARR), hydrogen bonds properties: planarity (number of multiple and aromatic bonds — nBM, nArCO, nArOR, ARR), forming ability (nOHp, nO, H-052, P_VSA_LogP_2, P_VSA_p_4, CATS2D_07_DA, CATS2D_04_AN), hydrogen bonds forming ability (nOHp, nO, H-052, P_VSA_LogP_2, P_VSA_p_4, CATS2D_07_DA, ionic properties (MATS7e, GATS5i, EE_B(s)). CATS2D_04_AN), ionic properties (MATS7e, GATS5i, EE_B(s)). 2.1.1. Consensus of Model Predictions Table 1. Statistic parameters of the best three single models and consensus classification models. Consensus modeling was performed by integrated predictions among two QSAR models (2-model Single Models 2-Model Consensus 3-Model Consensus consensus, three possible combinations) and all three models together (3-model consensus). In general, NN-C NN-D Q-D NN-C + NN-D NN-C + Q-D NN-C + Q-D A B A + B researchers acknowledged an increase in overall predictive performance, if consensus of predictions TP 44 44 46 33 33 41 29 19 48 among a battery of models was applied [69]. Following the protocol of Marzo et al. [69] we made TN 61 65 58 47 49 55 45 17 62 various combinations of consensus of model predictions by combining predictions from NN-C, FP 9 5 12 6 2 1 1 7 8 NN-D and Q-D models. The 3-model consensus A was the strictest and considered only predictions FN 6 6 4 0 2 0 0 2 2 with∑ all thr120 120 120 ee models (NN-C + NN-D +86 Q-D) in agreement. 86 In 3-model 97 consensus 75 45 ‘A + B’120 the PR 1 (0.98) 1 (0.92) 1 (0.97) 0.72 0.72 0.81 0.62 0.37 1 predictions with maximal one disagreement among the NN-C, NN-D and Q-D models were considered. SE 0.88 0.88 0.92 1.00 0.94 1.00 1.00 0.90 0.96 In 2-model consensus ‘NN-D + Q-D’ the predictions where both models were in agreement were SP 0.87 0.93 0.83 0.89 0.96 0.98 0.98 0.71 0.89 considered. In the same way the rest two combinations of a 2-model consensus (NN-C + NN-D and ACC 0.88 0.91 0.87 0.93 0.95 0.99 0.99 0.80 0.92 NN-C + Q-D) were also performed, but have lowest performance evaluation than ‘NN-D + Q-D’. NPV 0.91 0.92 0.94 1.00 0.96 1.00 1.00 0.89 0.97 The results of performance evaluation of consensus modeling are represented in Tables 1 and 2, and PPV 0.83 0.90 0.79 0.85 0.94 0.98 0.97 0.73 0.86 the detailed predictions in Table S1 (Supplementary Material). When compared with the single models, MCC 0.75 0.81 0.74 0.87 0.90 0.98 0.97 0.62 0.84 the accuracy performance of consensus of predictions was increased, but consequently the prediction A—complete agreement among predictions, B—maximal one disagreement among predictions. rate was decreased. The highest overall accuracy (0.99) was determined in 2-model consensus ‘NN-D + Q-D’ and 3-model consensus A, yet both consensus signiﬁcantly differ in prediction rate ability, 81% and 62%, respectively. Therefore, the best balance among accuracy and prediction rate was achieved with 3-model consensus ‘A + B’ (ACC = 0.92, PR = 100%). This 3-model consensus was shown as robust and reliable, and in evaluation parameters outperformed all single models. In this regard, the high impact of integration of predictions of three models was conformed and in concordance with expectations the overall predictive performance was increased. 2.1.2. Application of Classiﬁcation Models on Phenolic Dataset Developed QSAR models enabled efﬁcient application in screening processes regarding bioavailability of dietary/pharmaceutically interesting compounds. With single classiﬁcation models and 2-/3-model consensus of predictions we estimated the BTL transport potential for compounds, which were not experimentally tested for BTL transport activity (phenolic dataset, n = 300). Results of Molecules 2019, 24, x FOR PEER REVIEW 8 of 21 Among single models the highest number of active compounds was predicted with the model NN-D (138), which was significantly higher than in other models (NN-C = 75, Q-D = 72). However, none of single models or consensus of predictions did not recognize groups of phenols that are more likely to interact with BTL (Figure 5). For sure the most promising active compounds are those 15 that were predicted in all models: luteolin (ID4), kaempferol (ID86), eriodictyol (ID95), pinobanksin (ID117), cianidanol (ID127), leucodelphinidin (ID131), ellagic acid (ID181), rosmarinic acid (ID182), gallic acid (ID199), methyl gallate (ID200), 3-methoxy-4-hydroxybenzoic acid (ID209), 3-methoxy-4-hydroxyhippuric acid (ID211), decanyl caffeate (ID225), oleuropein (ID226), PACD3 (ID280) (see Table S2, column A, Supplementary Material). Since in vivo/in vitro experimental data on membrane transport ability for majority of the above mentioned compounds are not available either for BTL neither for other transporters from the list in Table 3 (except for four compounds and Molecules 2019, 24, 837 8 of 21 few transporters, see results in subsection 2.3.), the in silico predictions are currently the only relevant estimations of their membrane transport activities. Interestingly, based on our results BTL might be also involved in the gastric absorption of the flavone luteolin, as its gastric uptake was predictions are represented in Figure 5 and Table S2 (Supplementary Material). Consensus ‘A + B’ observed in vivo [70]. Indeed, this is in support of hypothesis of Passamonti et al. [35] that at the and single models N-C and Q-D performed with a 100% prediction rate with most of the compounds gastric level the BTL-mediated mechanism could play an important role in active transport of within AD (A + B = 300, NN-C = 283, Q-D = 278). On the other hand, the model NN-D resulted polyphenols. The isoflavone daizdein, on the other hand, was classified as inactive with our with a lower number of compounds in AD (NN-D = 208). As expected, lower prediction rate was consensus of QSAR models and was also experimentally identified as not being a BTL substrate [35]. evaluated for other consensus of predictions (NN-D + Q-D = 50%, A = 36%), due to strictest conditions. However, study of Piskula et al. [71] demonstrated that daizdein can be absorbed from the rat Generally, the integration of multiple models increased the overall reliability of predictions in all stomach, thus in its gastric uptake probably another specific OAT transporter, a monocarboxylate consensus combinations, also increased the prediction rate for phenolic compounds in consensus transporter, is involved [35]. ‘A + B’, but decreased in other consensus (NN-D + Q-D, A). Figure 5. Representation of classiﬁcation of 300 compounds with three separate classiﬁcation models Figure 5. Representation of classification of 300 compounds with three separate classification models (NN-C, NN-D, Q-D) and three consensus models (A + B, NN-D + Q-D, A) on graphical map. (NN-C, NN-D, Q-D) and three consensus models (A + B, NN-D + Q-D, A) on graphical map. Using in silico models one is challenged with the paradigm of selecting single model or very strict consensus (e.g., A) with high accuracy and narrow AD, or on the price of broadening of AD decide for wider consensus (e.g., A + B). In this regard, the number of active compounds predictions varied from 15 in consensus A to 65 in consensus ‘A + B’ (Table S2, Supplementary Material). Among single models the highest number of active compounds was predicted with the model NN-D (138), which was signiﬁcantly higher than in other models (NN-C = 75, Q-D = 72). However, none of single models or consensus of predictions did not recognize groups of phenols that are more likely to interact with BTL (Figure 5). For sure the most promising active compounds are those 15 that were predicted in all models: luteolin (ID4), kaempferol (ID86), eriodictyol (ID95), pinobanksin (ID117), cianidanol (ID127), leucodelphinidin (ID131), ellagic acid (ID181), rosmarinic acid (ID182), gallic acid (ID199), methyl gallate (ID200), 3-methoxy-4-hydroxybenzoic acid (ID209), 3-methoxy-4-hydroxyhippuric acid (ID211), decanyl caffeate (ID225), oleuropein (ID226), PACD3 (ID280) (see Table S2, column A, Supplementary Material). Since in vivo/in vitro experimental data on membrane transport ability for majority of the above mentioned compounds are not available either for BTL neither for other transporters from the list in Table 3 (except for four compounds and few transporters, see results in Section 2.3), the in silico predictions are currently the only relevant estimations of their membrane transport activities. Interestingly, based on our results BTL might be also involved in the gastric absorption of the ﬂavone luteolin, as its gastric uptake was observed in vivo [70]. Indeed, this is in support of hypothesis of Passamonti et al. [35] that at the gastric level the BTL-mediated mechanism could play an important role in active transport of polyphenols. The isoﬂavone daizdein, on the other hand, was classiﬁed as inactive with our consensus of QSAR models and was also experimentally identiﬁed as not being a BTL substrate [35]. However, study of Piskula et al. [71] demonstrated that daizdein can be absorbed from the rat stomach, thus in its gastric uptake probably another speciﬁc OAT transporter, a monocarboxylate transporter, is involved [35]. Molecules 2019, 24, 837 9 of 21 Table 3. List of transporters from Metrabase and ChemBench platforms. Protein Symbol Gene Symbol Name Tissue Expression Data M,C MDR1 ABCB1 ATP-binding cassette sub-family B member 1, P-glycoprotein intestine, kidney, liver, brain BSEP ABCB11 ATP binding cassette subfamily B member 11 liver M,C MRP1 ABCC1 ATP-binding cassette, sub-family C member 1 intestine, kidney, liver, brain M,C ABCC2 ATP-binding cassette sub-family C member 2 stomach, intestine, kidney, liver, brain MRP2 M,C MRP3 ABCC3 ATP-binding cassette sub-family C member 3 stomach, intestine, kidney, liver, brain M,C ABCC4 ATP-binding cassette, sub-family C member 4 stomach, intestine, kidney, liver, brain MRP4 MRP5 ABCC5 ATP Binding Cassette Subfamily C member 5 kidney, brain M,C BCRP1 ABCG2 ATP-binding cassette sub-family G member 2 stomach, intestine, kidney, liver, brain SLCO1A2 solute carrier organic anion transporter family member 1A2 kidney, liver, brain OATP1A2 OATP2A1 SLCO2A1 solute carrier organic anion transporter family member 2A1 intestine, kidney, liver, brain OATP1B1 SLCO1B1 solute carrier organic anion transporter family member 1B1 liver OATP1B3 SLCO1B3 solute carrier organic anion transporter family member 1B3 liver M,C SLCO2B1 solute carrier organic anion transporter family member 2B1 intestine, kidney, liver, brain OATP2B1 OATP3A1 SLCO3A1 solute carrier organic anion transporter family member 3A1 intestine, kidney, liver, brain OATP4A1 SLCO4A1 solute carrier organic anion transporter family member 4A1 stomach, intestine, kidney, liver, brain GLUT1 SLC2A1 solute carrier family 2 (facilitated glucose transporter) member 1 intestine, kidney, liver, brain NTCP SLC10A1 solute carrier family 10 member 1 liver ASBT SLC10A2 solute carrier family 10 Member 2 intestine SLC15A1 solute carrier family 15 member 1 intestine PEPT1 M,C MCT1 SLC16A1 solute carrier family 16 member 1, monocarboxylic acid transporter 1 stomach, intestine, kidney, liver, brain M,C OCT1 SLC22A1 solute carrier family 22 (organic cation transporter) member 1 intestine, kidney, liver BTL * / bilitranslocase, bilirubin membrane transporter stomach, intestine, kidney, liver * new, not available in Metrabase and ChemBench, M—available in Metrabase, C—available in ChemBench. Molecules 2019, 24, 837 10 of 21 2.2. BTL Regression Models The activity predictions for the BTL inhibition constants pKi [mmol/L] were performed with the best in-house predictive CP-ANN models (M1–M4) [59]. The neural networks were selected regarding results from the previous studies [59]. Firstly, we analyzed, if the regression models were accurately developed and validated regarding the need of evenly frequency distribution of compounds in all ranges of BTL biological activity in the training and validation datasets. This condition was fully satisﬁed and a compounds frequency distribution in ﬁve intervals of BTL activity is shown in Figure S1 (Supplementary Material). For further detailed evaluation of models a ﬁve quality parameters of 2 2 2 validation were employed (R , RMSE, Q , CCC, r ), and their performance is represented in Table 4. F3 m Statistical differences among models are low and in general all validation parameters characterize model M1 as the best one. The validation parameter r was specially reported to help in identifying the best models among a set of comparable models [72], but in this case all parameters were equally sensitive. Since GA approach was used in the development and optimization of models, consequently a large availability of different models predicting the same endpoint was produced. Thus, to avoid having cluster of similar models we tested the diversity of the top scoring models by measuring of Hamming distance between them. The results are shown in Table S5 (Supplementary Material). The distances among models are in range of 7–17. This indicated that numbers of variables not shared by the models are high. Accordingly, our set of top models was shown to be diverse and consensus analyses were justiﬁed. An average response and weighted average response were used as proposed by Todeschini et al. [65]. The consensus gave better statistical ﬁts of predictions in regard to the single models. Table 4. Statistic parameters of the best four regression models. 2 2 2 2 2 2 Model R RMSE R Q CCC R Q CCC RMSE CCC r tr ev ev m iv iv F3iv iv F3ev all all (all) M1 0.996 0.266 0.949 0.957 0.970 0.872 0.927 0.939 0.198 0.987 0.821 M2 0.999 0.280 0.943 0.953 0.968 0.842 0.909 0.927 0.206 0.986 0.812 M3 0.998 0.289 0.939 0.950 0.964 0.839 0.908 0.934 0.213 0.985 0.806 M4 0.995 0.295 0.937 0.948 0.964 0.760 0.863 0.899 0.243 0.980 0.776 Application of Regression Models on Phenolic Dataset Predictions were made for 205 compounds. These were compounds that were classiﬁed as active with at least one model (Table S7, Supplementary Material). In order to produce more reliable estimates of our endpoint the consensus modeling strategy was also applied. Most of models were well applicable for phenols, since majority of compounds were in AD of models (>75%); exception was model M3 with only 50% of compounds in AD. Moreover, the predictions were highly harmonized, since varied for less than 10% of standard deviation, if compared among all models (M1-M4). Up to our knowledge only one study exist on extrapolation of BTL models to make predictions of new compounds, which were not included in model development or validation [59]. Comparison with that study revealed that BTL models have wider capacity for prediction of phenolic compounds than other antioxidants and antiprions (Table S6, Supplementary Material). 2.3. Metrabase and Chembench Analyses Based on data and tools available on Metrabase and Chembench platforms we made the comprehensive virtual proﬁling of membrane transport interactions for compounds of both datasets. Herein, we performed the unique transporter proﬁle of studied compounds, which is based on the results from new BTL models and hits/predictions from over 20 membrane transporters of Metrabase and Chembench (Figures 6 and 7). Molecules 2019, 24, x FOR PEER REVIEW 11 of 21 37–84), nucleobases, nucleosides and nucleotides (ID 1–36). The only exception was few flavonoids (ID 65–70). On the other hand, drugs and endogenous substances (ID 85–120), which are of interest for pharmaceutical industry, were better studied and experimentally analyzed, therefore more hits Molecules 2019, 24, 837 11 of 21 were found. non-inhibitor inhibito ambigious expression MDR1 MRP1 MRP2 MRP3 MRP4 BCRP1 GLUT1 MCT1 OCT1 OATP1A2 OATP1B1 OATP1B3 OATP2A1 OATP2B1 OATP3A1 OATP4A1 BTL nucleobase nucleoside nucleotide phenols drug other Figure 6. Activity proﬁle for 120 compounds including BTL experimental data [53,54] and data Figure 6. Activity profile for 120 compounds including BTL experimental data [53,54] and data Molecules 2019, 24, x FOR PEER REVIEW 12 of 21 available for 16 transporters in Metrabase [46]. Grey—no data available in database. available for 16 transporters in Metrabase [46]. Grey—no data available in database. Based on these results we can conclude that for phenolic compounds, nucleobases, nucleosides and nucleotides the only available information regarding experimental transport activity of the membrane transporters is those of BTL provided from research group of prof. Passamonti [53,54], which are not included in Metrabase. Further on, we made analyses in Metrabase for compounds of consensus A, which were predictive as active (15 compounds, see Subsection 2.1.2). We found out that only for luteolin (ID4), kampferol (ID86), cianidanol (ID127) and eriodictyol (ID95) a few experimental studies exist in this database; luteolin (MDR1–inhibitor, BCRP1–inhibitor, MRP2– non-inhibitor), kaempferol (BCRP1–inhibitor, MRP2–non-inhibitor, OCT1–inhibitor, OATP1A2– inhibitor), cianidanol (MRP2–non-inhibitor or substrate, OCT1–non-inhibitor, OATP2B1–inhibitor or non-inhibitor) eriodictyol (MRP2–non-inhibitor). Besides this four compounds, which were captured the significant attention of the scientific audience for their potential anti-oxidative, anti-tumor, and anti-inflammatory effects, also other phenolic compound of natural origin like oleuropein (ID226) and leucodelphinidin (ID131) are of interest due to their anti-neurodegenerative and hypolipidaemic effects [73,74]. Thus, the predictions about transport ability by BTL and 14 others importers and exporters that are represented in Tables S2 and S9 (Supplementary Material) are of high interest for clarification about their active transport across membrane. Further on, we made predictions for 420 compounds (BTL and phenolic datasets) by using QSAR models of 14 transporters available in Human intestinal transporter database at Chembench platform [44]. The results of predictions are represented in Figure 7, Table S8 and S9 (Supplementary Material). In general, the models have been found to have good prediction ability and coverage for phenolic compounds, since only few compounds were out of AD or unable to predict (Table S9, Supplementary Material). Yet, almost all models had poor prediction ability for nucleobases, nucleosides and nucleotides (ID 1–36, Table S8, Supplementary Material). Poor prediction ability was observed also for some flavonoids in models of MRP4, MRP5, MCT1, NTCP, ASBT and OCT1 (anthocyanidins (ID 71–84) and flavonols (ID 52–64)). Figure 7. Activity proﬁle of computational predictions for 14 transporters from ChemBench and for Figure 7. Activity profile of computational predictions for 14 transporters from ChemBench and for our BTL model, red—active, blue—inactive, grey—no prediction; (a) BTL dataset of 120 compounds, our BTL model, red—active, blue—inactive, grey—no prediction; (a) BTL dataset of 120 compounds, (b) phenols dataset of 300 compounds. (b) phenols dataset of 300 compounds. We compared the transport activity data among BTL and 16 transporters that are publicly 3. Discussion available in Metrabase (cheminformatics and bioinformatics database) [46]. Analysis of hits of BTL In spite a lack of adequate experimental validation for our in silico predictions, some dataset (compounds from Table S1, Supplementary material) in Metrabase database is represented information could be gained by comparing structurally similar compounds. Among the fifteen the on Figure 6. It was found out that almost no experimental data exist for the phenolic compounds most promising active compounds, which were selected with the BTL QSAR models, luteolin, kaempferol, eriodictyol and pinobanksin were found to share the highest structural similarity with quercetin. Quercetin, malvidin 3-glucoside, bilirubin and bromosulfophthalein are up to date the only compounds identified experimentally as BTL substrates [55,75]. Hopefully, the results published in this study would be inspiration for in vitro experts to test the most promising predicted active compounds with the BTL transport activity assay. The compounds of natural origin with high health benefit should be selected first, like oleuropein, luteolin, eriodictyol, cianidanol, and leucodelphinidin. The clinic reports highlighted the impact of knowledge about how various therapeutics or dietary supplements interacts with membrane transporters [76]. Herein, for the first time were represented joint analyses of transport activities of BTL and several others membrane transporters. Results not solely present a new insight in interactions between phenols and several membrane transporters, but can help also in assessment of their potential absorption and disposition properties, yet gain also valuable information for understanding possible interference of phenols with the transport activity of drugs. Since ingestion of extra dietary phenolic supplements as support to chemotherapy was proposed as promising therapeutic approach [7,10], but such diet could consequently influence drugs absorption and disposition [58]. For proper drug application this is for Molecules 2019, 24, 837 12 of 21 (ID 37–84), nucleobases, nucleosides and nucleotides (ID 1–36). The only exception was few ﬂavonoids (ID 65–70). On the other hand, drugs and endogenous substances (ID 85–120), which are of interest for pharmaceutical industry, were better studied and experimentally analyzed, therefore more hits were found. Based on these results we can conclude that for phenolic compounds, nucleobases, nucleosides and nucleotides the only available information regarding experimental transport activity of the membrane transporters is those of BTL provided from research group of prof. Passamonti [53,54], which are not included in Metrabase. Further on, we made analyses in Metrabase for compounds of consensus A, which were predictive as active (15 compounds, see Section 2.1.2). We found out that only for luteolin (ID4), kampferol (ID86), cianidanol (ID127) and eriodictyol (ID95) a few experimental studies exist in this database; luteolin (MDR1–inhibitor, BCRP1–inhibitor, MRP2–non-inhibitor), kaempferol (BCRP1–inhibitor, MRP2–non-inhibitor, OCT1–inhibitor, OATP1A2–inhibitor), cianidanol (MRP2–non-inhibitor or substrate, OCT1–non-inhibitor, OATP2B1–inhibitor or non-inhibitor) eriodictyol (MRP2–non-inhibitor). Besides this four compounds, which were captured the signiﬁcant attention of the scientiﬁc audience for their potential anti-oxidative, anti-tumor, and anti-inﬂammatory effects, also other phenolic compound of natural origin like oleuropein (ID226) and leucodelphinidin (ID131) are of interest due to their anti-neurodegenerative and hypolipidaemic effects [73,74]. Thus, the predictions about transport ability by BTL and 14 others importers and exporters that are represented in Tables S2 and S9 (Supplementary Material) are of high interest for clariﬁcation about their active transport across membrane. Further on, we made predictions for 420 compounds (BTL and phenolic datasets) by using QSAR models of 14 transporters available in Human intestinal transporter database at Chembench platform [44]. The results of predictions are represented in Figure 7, Tables S8 and S9 (Supplementary Material). In general, the models have been found to have good prediction ability and coverage for phenolic compounds, since only few compounds were out of AD or unable to predict (Table S9, Supplementary Material). Yet, almost all models had poor prediction ability for nucleobases, nucleosides and nucleotides (ID 1–36, Table S8, Supplementary Material). Poor prediction ability was observed also for some ﬂavonoids in models of MRP4, MRP5, MCT1, NTCP, ASBT and OCT1 (anthocyanidins (ID 71–84) and ﬂavonols (ID 52–64)). 3. Discussion In spite a lack of adequate experimental validation for our in silico predictions, some information could be gained by comparing structurally similar compounds. Among the ﬁfteen the most promising active compounds, which were selected with the BTL QSAR models, luteolin, kaempferol, eriodictyol and pinobanksin were found to share the highest structural similarity with quercetin. Quercetin, malvidin 3-glucoside, bilirubin and bromosulfophthalein are up to date the only compounds identiﬁed experimentally as BTL substrates [55,75]. Hopefully, the results published in this study would be inspiration for in vitro experts to test the most promising predicted active compounds with the BTL transport activity assay. The compounds of natural origin with high health beneﬁt should be selected ﬁrst, like oleuropein, luteolin, eriodictyol, cianidanol, and leucodelphinidin. The clinic reports highlighted the impact of knowledge about how various therapeutics or dietary supplements interacts with membrane transporters [76]. Herein, for the ﬁrst time were represented joint analyses of transport activities of BTL and several others membrane transporters. Results not solely present a new insight in interactions between phenols and several membrane transporters, but can help also in assessment of their potential absorption and disposition properties, yet gain also valuable information for understanding possible interference of phenols with the transport activity of drugs. Since ingestion of extra dietary phenolic supplements as support to chemotherapy was proposed as promising therapeutic approach [7,10], but such diet could consequently inﬂuence drugs absorption and disposition [58]. For proper drug application this is for sure of high importance. Moreover, results of the systematic analysis performed in this study are also important for elucidation of functional role of BTL, as up to our knowledge only limited data exist. Up to date, only one Molecules 2019, 24, 837 13 of 21 synopsis of reported interactions with BTL and other membrane transporters one a few compounds (>50) is available [58]. Since in general not much is known about interactions between phenols and transporters, the results from this study are of interest [77]. Moreover, we managed to conﬁrm our expectations, that newly developed BTL QSAR models have good predictive ability and coverage for phenolic compounds. In future it would be of interest to make a similar study on the pharmaceutically interested nucleobase derivatives [78–80], for which it is assumed that BTL could have important role on their transport [54]. More than half a century has passed since an implementation of Hansch-Fujita approach in the formulation of QSAR, which provided a new perspective for chemical-biological interactions and a number of successful applications are known especially in drug discovery [77]. Nowadays, we have abundance of QSAR models for predictions of numerous endpoints in various ﬁelds of science. In this study we tried to collect and compare the QSAR models of membrane transporters to make a comprehensive transporter proﬁles for purposes of virtual screening of compounds for their bioavailabilty. Now point of future investigation should be to switch in new level beyond individual approaches and integrate various QSAR analyses with ontology and omics data [77,81–83]. Such simultaneously predicting with joint data analyses are shown to increase the accuracy of models by exploiting their common representation and identifying common features between individual properties, which are frequently strongly correlated with one another [81,83]. For sure not in so far future a chemical characterization will be even more empowered with an artiﬁcial intelligence assisted software, which would make decisions based on integration of numerous data from in silico, in vitro, in vivo and in situ analyses [84]. Thus, hopefully models from this study will serve like small piece of a large puzzle in the complex biochemical system analysis. 4. Materials and Methods Quantitative structure-activity relationships (QSARs) are widely used in silico methods to predict biological activity for untested chemicals (Figure 1). This is proven, time and cost-effective approach, which is in most cases sufﬁcient alternative method to animal testing [44]. For building and validation of models the appropriate statistical algorithms and the data matrix which includes numerical values of chemical structures and empirical values of biological activity are needed. 4.1. Dataset The dataset for modeling included 120 aromatic organic compounds (43 phenols and 77 various aromatic compounds like nucleobase derivatives, drugs, dyes, etc.) [59]. In general chemical structures of all compounds were optimized and numerically coded with molecular descriptors (MD). The MD calculation and selection of inﬂuential MD is described in the Section 4.2.1. The corresponding empirical data of BTL transport activity assay were collected from experiments of Župerl et al. [54] and Karawajczk et al. [53]. Inhibition constants (K ) were determined based on Michaelis-Menten constants, which were obtained from spectrophotometrical evaluation of BTL transport activity on selected compounds in medium of sulfobromophtalein (BSP) and rat liver plasma membrane vesicles. According to threshold criteria proposed by Župerl et al. [54] the dataset composed of 50 active compounds (inhibitors, pK > 1.3) and 70 inactive compounds (non-inhibitors, pK 1.3). The compounds are listed in Table S1 (Supplementary Material). The test dataset was compiled from literature. It encompasses natural occurring and synthetic phenolic compounds having various potential health beneﬁts. Altogether 300 compounds were collected from over 30 publications [7,11–30,85–93], see Table S2 (Supplementary Material). Compounds were arranged according to their structural similarity into 12 groups: ﬂavones/isoﬂavones, ﬂavonols, ﬂavanones/ isoﬂavanones, ﬂavanonols, ﬂavanols, xanthones, chromones, coumarins, chalcones, phenolic acid derivatives, simple phenols and other compounds. Experimental evaluation of potential role of selected phenolic compounds in preventing various chronic diseases are explained in detail in the references listed in Table S2 (Supplementary Material). Molecules 2019, 24, 837 14 of 21 4.2. Computational Analysis 4.2.1. Molecular Descriptors Molecular structures were represented with a set of molecular descriptors (MDs). First, the MOPAC software [94] was used for optimization of the 2D chemical structures into 3D conformations considering the minimal energy criterion for the optimal conformation of the compound in vacuum. Then numerical coding of 3D conformations was performed with MDs, which were calculated with the CODESSA program [95]. With the applied software more than 400 MDs for each compound were calculated. Second, the numerical encoding of 2D conformations was performed with Dragon 7.0 software (Kode srl., Pisa, Italy) [96], which calculated over 3000 MDs for each molecule. The dataset with Codessa and Dragon descriptors were further on treated separately. So, obtained two datasets were normalized to zero mean and unit standard deviation for each descriptor. Afterwards, the reduction of numerous descriptors was performed by omitting the MDs with zero variance or high intercorrelation (>0.9). The set of descriptors was then reduced according to the similarity criterion in the Kohonen map by following the protocol of Martinci ˇ c ˇ et al. [59], which resulted in the ﬁnal data pool of 66 and 87 MDs for Codessa and Dragon dataset, respectively. 4.2.2. Generation of Models The inputs for model building were various m-dimensional vectors representing the chemical structure, m being the number of MDs selected (independent variables), and the targets (properties, i.e. dependent variables) corresponding to pK of BTL transport activity. We used CPANNatNIC software for the splitting of data for modeling into training and validation set [59,92]. For model ﬁtting the linear and nonlinear regression approaches were used. The multiple linear regression models were developed with Qsarins software (DiSTA, Varese, Italy, www.qsar.it) [68]. The supervised learning models were developed by using counter-propagation artiﬁcial neural network (CP-ANN) algorithm and in-house software [97,98]. For classiﬁcation and prediction of unknown compounds two QSAR modeling strategies were adopted. Firstly, the classiﬁcation models were developed to separate chemicals into active/non-active classes, and then optimized regression models were used for the active chemicals to predict the values of the inhibition constants [59]. All 120 compounds were considered for the construction and optimization of the classiﬁcation model, while the predictive models are based on active compounds only. For optimization we applied cross-validation (CV) and genetic algorithm (GA) [65,98] in order to select the inﬂuential descriptors and to improve the predictive ability and robustness of the models. In models selection the applicability domain (AD) [67] was applied to evaluate the model predictions within the established chemical space limits. 4.2.3. Statistical Evaluation of Models The criteria for selection of the best classiﬁcation models, so called quality indicators, are listed in Table 5 [99,100]. Models with the highest quality indicators of the validation set and wider applicability (larger AD) were further analyzed. Two approaches were used for AD evaluation, the cumulative distributions of Euclidian distances to central neurons (MEDS) for CP-ANN models [67] and the leverage values for MLR models [68]. Several models were built by each of the selected modelling method. Further on, the consensus of model predictions was performed by combining predictions of three or two QSAR models. The consensus was performed following the protocol of Marzo et al. [69], which including arranging of compounds in two types: A—complete agreement among predictions, B-maximum one disagreement among the predictions. Molecules 2019, 24, 837 15 of 21 Table 5. Quality indicators for evaluation of different classiﬁcation models. Acronym Quality Indicator Formula TP true positive TN true negative FP false positive FN false negative PR predict ratio predicted compounds/total compounds SE, TPR sensitivity, true positive rate TP/(TP + FN) SP, TNR speciﬁcity, true negative rate TN/(TN + FP) ACC accuracy (SP + SE)/2 NPV negative predictive value TN/(TN + FN) PPV positive predictive value, precision TP/(TP + FP) ((TP*TN)-(FP*FN))/ [(TP + FP)(TP + MCC Matthews correlation coefﬁcient FN)(TN + FP)(TN + FN)] The criteria for selection of the best regression models were several validation parameters 2 2 2 (R , RMSE, CCC, Q , r ) [64,70]. The consensus approach was also applied for the best four F3 m regression models (M1-M4) by using the weighted average response. The calculation was performed following the Equation (1); M—number of models, y —response estimated by the k-th model, h —leverage [65]: M k k=1 h y = (1) M 1 k=1 In order to evaluate diversity of models the Hamming distances d between models were accounted based on the two-way table [65] (Equation (2)). The table represent the relationship between two binary vectors, each vector represent one model; b—the number of cases that for same position there is 1 in vector I and 0 vector II, c—the number of cases that for same position there is 0 in vector I and 1 vector II Values in Table S5 (Supplementary Material) represent the number of MD not shared among models: d = b + c (2) 4.3. Metrabase and Chembench Analyses The public available cheminformatics and bioinformatics database for small molecule transporter data analysis (Metrabase) [46,101] was screened for transport activity proﬁle of 120 compounds (BTL dataset). The synopsis of transport activity among BTL and 16 transporters from Metrabase was performed. The selected 16 transporters are listed in Table 3. Further on, in Human intestinal transporter database, available on Chembench platform, the QSAR models of 14 transporters were used to make predictions for 420 compounds (BTL and phenolic datasets) [44,102]. The transporters are listed in Table 3 and detailed list of QSAR models properties is summarized in Table S3 (Supplementary Material). The comparison of transport activity among BTL and 14 transporters from Chembench was performed for both datasets. 5. Conclusions New classiﬁers for BTL transport activity were developed. Based on in silico approach for estimation of BTL transport activity the classiﬁcation for dietary/pharmaceutically interesting compounds (natural dietary phenols and synthetic phenol-like drugs) were performed. Using QSAR classiﬁcation models (NN-C, NN-D, Q-D) we estimated the BTL transport activity of 300 compounds, which have not been previously experimentally evaluated. Moreover, the consensus among battery of models was applied, since the detailed evaluation of models shown better performance than individual models. With various predictive CP-ANN models (M1-M4) then the transport activity prediction (inhibition constant pKi [mmol/L]) for the compounds classiﬁed as active was assessed. The reliability Molecules 2019, 24, 837 16 of 21 of all predictions was evaluated with the applicability domain (AD) of each individual classiﬁcation and regression model. In general, BTL is known as a potential delivery system for wide variety of the vitally important endogenous substrates and drugs, yet with the models represented in this study a good ability for predicting phenolic compounds was shown too. The 205 compounds from the dataset of 300 phenols have a potential to interact with BTL. The central part of this study was the synopsis of unique transporter proﬁles of BTL in comparison with 20 other membrane transporters from Metrabase and Chembench. Results present a new insight in interactions between phenols and several membrane transporters, which can help not solely in assessment of their potential absorption and disposition properties but also gain insight in possible interference of phenols with the transport activity of drugs, what can be valuable information in drug application. Supplementary Materials: The list of ChemBench models and detailed list of BTL and phenolic datasets with extended results of data analyses (Tables S1–S9) are available as supplementary material online. Author Contributions: Conceptualization and design of study, K.V. and M.N.; data curation, K.V.; calculations, data analysis and validation, K.V.; writing—original draft preparation and editing, K.V.; writing—review, discussion, M.N.; project administration, M.N.; funding acquisition, M.N. Funding: This work was ﬁnancially supported by the grants from the Slovenian Research Agency (national program P1-0017 and research project J3-8209). Acknowledgments: We thank Viktor Drgan for technical support and Paola Gramatica for availability of Qsarins license. Conﬂicts of Interest: The authors declare no conﬂict of interest. References 1. Del Rio, D.; Rodriguez-Mateos, A.; Spencer, J.P.; Tognolini, M.; Borges, G.; Crozier, A. Dietary (poly)phenolics in human health: Structures, bioavailability and evidence of protective effects against chronic diseases. Antioxid. Redox. Signal. 2013, 18, 1818–1892. [CrossRef] [PubMed] 2. Dai, J.; Mumper, R.J. Plant Phenolics: Extraction, Analysis and Their Antioxidant and Anticancer Properties. Molecules 2010, 15, 7313–7352. [CrossRef] [PubMed] 3. Harborne, J.B. Plant phenolics. In Encyclopedia of Plant Physiology, Volume 8: Secondary Plant Products; Bell, E.A., Charlwood, B.V., Eds.; Springer: Berlin, Germany; New York, NY, USA, 1980; pp. 329–395. 4. Khoddami, A.; Wilkes, M.A.; Roberts, T.H. Techniques for analysis of plant phenolic compounds. Molecules 2013, 18, 2328–2375. [CrossRef] [PubMed] 5. Nicholson, R.L.; Vermerris, W. Phenolic compound biochemistry; Springer: Berlin, Germany, 2006; pp. 107–108. 6. FooDB version 1.0. Available online: www.foodb.ca (accessed on 15 October 2018). 7. Fresco, P.; Borges, F.; Diniz, C.; Marques, M.P. New Insights on the Anticancer Properties of Dietary Polyphenols. Med. Res. Reviews 2006, 26, 747–766. [CrossRef] [PubMed] 8. Pietta, P.G. Flavonoids as antioxidants. J. Nat. Prod. 2000, 63, 1035–1042. [CrossRef] [PubMed] 9. Lu, Z.; Nie, G.; Belton, P.S.; Tang, H.; Zhao, B. Structure-activity relationship analysis of antioxidant ability and neuroprotective effect of gallic acid derivatives. Neurochem. Internat. 2006, 48, 263–274. [CrossRef] [PubMed] 10. Rasines-Perea, Z.; Teissedre, P.L. Grape Polyphenols’ Effects in Human Cardiovascular Diseases and Diabetes. Molecules 2017, 22, 68. [CrossRef] [PubMed] 11. Khalil, M.I.; Sulaiman, S.A. The potential role of honey and its polyphenols in preventing heart diseases: A review. Afr. J. Tradit. Complement. Altern. Med. 2010, 7, 315–321. [CrossRef] [PubMed] 12. Machado, N.F.L.; Calheiros, R.; Fiuza, S.M.; Borges, F.; Gaspar, A.; Garrido, J.; Marques, M.P. Phenolic esters with potential anticancer activity—The structural variable. J. Mol. Model. 2007, 13, 865–877. [CrossRef] [PubMed] 13. Bauvois, B.; Puiffe, M.L.; Bongui, J.B.; Paillat, S.; Monneret, C.; Dauzonne, D. Synthesis and Biological Evaluation of Novel Flavone-8-acetic Acid Derivatives as Reversible Inhibitors of Aminopeptidase N/CD13. J. Med. Chem. 2003, 46, 3900–3913. [CrossRef] [PubMed] Molecules 2019, 24, 837 17 of 21 14. Locatelli, C.; Rosso, R.; Santos-Silva, M.C.; de Souza, C.A.; Licínio, M.A.; Leal, P.; Bazzo, M.L.; Yunes, R.A.; Creczynski-Pasa, T.B. Ester derivatives of gallic acid with potential toxicity toward L1210 leukemia cells. Bioorg. Med. Chem. 2008, 16, 3791–3799. [CrossRef] [PubMed] 15. Fernandes, I.; Pérez-Gregorio, R.; Soares, S.; Mateus, N.; de Freitas, V. Wine Flavonoids in Health and Disease Prevention. Molecules 2017, 22, 292. [CrossRef] [PubMed] 16. Fiuza, S.M.; Gomes, C.; Teixeira, L.J.; Girão da Cruz, M.T.; Cordeiro, M.N.D.S.; Milhazes, N.; Borges, F.; Marques, M.P.M. Phenolic acid derivatives with potential anticancer properties—A structure-activity relationship study. Part 1: Methyl. propyl and octyl esters of caffeic and gallic acids. Bioorg. Med. Chem. 2004, 12, 3581–3589. [CrossRef] [PubMed] 17. Sergediene, ˙ E.; Jönsson, K.; Szymusiak, H.; Tyrakowska, B.; Rietjens, I.M.C.M.; Cenas, ˙ N. Prooxidant toxicity of polyphenolic antioxidants to HL-60 cells: Description of quantitative structure-activity relationships. FEBS Lett. 1999, 462, 392–396. [CrossRef] ˇ ˇ 18. Nemeikaite- ˙ Cenien ˙ e, ˙ A.; Imbrasaite, ˙ A.; Sergediene, ˙ E.; Cenas, ˙ N. Quantitative structure-activity relationships in prooxidant cytotoxicity of polyphenols: Role of potential of phenoxyl radical/phenol redox couple. Archives Biochem. Biophys. 2005, 441, 182–190. [CrossRef] [PubMed] 19. Gomes, C.A.; da Cruz, T.G.; Andrade, J.L.; Milhazes, N.; Borges, F.; Marques, M.P.M. Anticancer activity of phenolic acids of natural or synthetic origin: A structure-activity study. J. Med. Chem. 2003, 46, 5395–5401. [CrossRef] [PubMed] 20. Nagaoka, T.; Banskota, A.H.; Tezuka, Y.; Saiki, I.; Kadota, S. Selective antiproliferative activity of caffeic acid phenethyl aster analogues on highly liver-metastatic murine colon 26-L5 carcinoma cell line. Bioorg. Med. Chem. 2002, 10, 3351–3359. [CrossRef] 21. Blank, V.C.; Poli, C.; Marder, M.; Roguin, L.P. 2004 Antiproliferative activity of various ﬂavonoids and related compounds: Additive effect of interferon-2b. Bioorg. Med. Chem. Lett. 2004, 14, 133–136. [CrossRef] [PubMed] 22. Chun, K.; Park, S.K.; Kim, H.M.; Choi, Y.; Kim, M.H.; Park, C.H.; Joe, B.Y.; Chun, T.G.; Choi, H.M.; Lee, H.Y.; et al. Chromen-based TNF- converting enzyme (TACE) inhibitors: Design synthesis and biological evaluation. Bioorg. Med. Chem. 2008, 16, 530–535. [CrossRef] [PubMed] 23. Lee, Y.T.; Don, M.J.; Liao, C.H.; Chiou, H.W.; Chen, C.F.; Ho, L.K. Effects of phenolic acid esters and amides on stimulus-induced reactive oxygen species production in human neutrophils. Clin. Chim. Acta 2005, 352, 135–141. [CrossRef] [PubMed] 24. Rao, Y.K.; Fang, S.H.; Tzeng, Y.M. Differential effects of synthesized 2 -oxygenated chalcone derivatives: Modulation of human cell cycle phase distribution. Bioorg. Med. Chem. 2004, 12, 2679–2686. [CrossRef] [PubMed] 25. Rao, K.Y.; Fang, S.H.; Tzeng, Y.M. Synthesis, growth inhibition, and cell cycle evaluations of novel ﬂavonoid derivatives. Bioorg. Med. Chem. 2005, 13, 6850–6855. [CrossRef] [PubMed] 26. Fang, S.H.; Rao, Y.K.; Tzeng, Y.M. Inhibitory effects of ﬂavonol glycosides from Cinnamomum osmophloeum on inﬂammatory mediators in LPS/IFN- -activated murine macrophages. Bioorg. Med. Chem. 2005, 13, 2381–2388. [CrossRef] [PubMed] 27. Benavente-García, O.; Castillo, J. Update on uses and properties of citrus ﬂavonoids: New ﬁndings in anticancer, cardiovascular, and anti-inﬂammatory activity. J. Agric. Food Chem. 2008, 56, 6185–6205. [CrossRef] [PubMed] 28. Cardenas, M.; Marder, M.; Blank, V.C.; Roguin, L.P. Antitumor activity of some natural ﬂavonoids and derivatives on various human and maurine cancer cell lines. Bioorg. Med. Chem. 2006, 14, 2966–2971. [CrossRef] [PubMed] 29. Bendini, A.; Cerretani, L.; Carrasco-Pancorbo, A.; Gómez-Caravaca, A.M.; Segura-Carretero, A.; Fernández-Gutiérrez, A.; Lercker, G. Phenolic Molecules in Virgin Olive Oils: A Survey of Their Sensory Properties, Health Effects, Antioxidant Activity and Analytical Methods. An Overview of the Last Decade. Molecules 2007, 12, 1679–1719. [CrossRef] [PubMed] 30. Seraﬁm, T.L.; Carvalho, F.S.; Maeques, M.P.M.; Chaleiros, R.; Silva, T.; Garrido, J.; Milhazes, N.; Borges, F.; Roleira, F.; Silva, E.T.; et al. Lipophilic caffeic and ferulic acid derivatives presenting cytotoxicity against human breast cancer cells. Chem. Res. Toxicol. 2011, 24, 763–774. [CrossRef] [PubMed] Molecules 2019, 24, 837 18 of 21 31. Li, F.; Wu, J.-J.; Wang, J.; Yang, X.-L.; Cai, P.; Liu, Q.-H.; Kong, L.-Y.; Wang, X.-B. Synthesis and pharmacological evaluation of novel chromone derivatives as balanced multifunctional agents against Alzheimer ’s disease. Bioorg. Med. Chem. 2017, 25, 3815–3826. [CrossRef] [PubMed] 32. Marsh, D.T.; Das, S.; Ridell, J.; Smid, S.D. Structure-activity relationships for ﬂavone interactions with 0 0 0 amyloid reveal a novel anti-aggregatory and neuroprotective effect of 2 ,3 ,4 -trihydroxyﬂavone (2-D08). Bioorg. Med. Chem. 2017, 25, 3827–3834. [CrossRef] [PubMed] 33. Letenneur, L.; Proust-Lima, C.; Le Gouge, A.; Dartigues, J.F.; Barberger-Gateau, P. Flavonoid intake and cognitive decline over a 10-year period. Am. J. Epidemiol. 2007, 165, 1364–1371. [CrossRef] [PubMed] 34. Manach, C.; Williamson, G.; Morand, C.; Scalbert, A.; Ramesy, C. Bioavailability and bioefﬁcacy of polyphenols in humans. Review of 97 bioavailability studies. Am. J. Clin. Nutr. 2005, 81, 230S–242S. [CrossRef] [PubMed] 35. Passamonti, S.; Terdoslavich, M.; Franca, R.; Vanzo, A.; Tramer, F.; Braidot, E.; Petrussa, E.; Vianello, A. Bioavailability of ﬂavonoids: A review of their membrane transport and the function of bilitranslocase in animal and plant organisms. Curr. Drug Metabol. 2009, 10, 369–394. [CrossRef] 36. Mennen, L.I.; Sapinho, D.; Ito, H.; Galan, P.; Hercberg, S.; Scalbert, A. Urinary excretion of 13 dietary ﬂavonoids and phenolic acids in free-living healthy subjects - variability and possible use as biomarkers of polyphenol intake. Eur. J. Clin. Nutr. 2008, 62, 519–525. [CrossRef] [PubMed] 37. Xie, H. Activity assay of membrane transport proteins. Acta Biochim. Biophys. Sin. 2008, 40, 269–277. [CrossRef] [PubMed] 38. Billat, P.A.; Roger, A.; Faure, S.; Lagarce, F. Models for drug absorption from the small intestine: Where are we and where are we going? Drug Discovery Today 2017, 22, 761–775. [CrossRef] [PubMed] 39. De Bruyn, T.; van Westen, G.J.P.; Jzerman, A.P.I.; Stieger, B.; de Witte, P.; Augustijns, P.F.; Annaert, P.P. Structure-Based Identiﬁcation of OATP1B1/3 Inhibitors. Mol. Pharmacol. 2013, 83, 1257–1267. [CrossRef] [PubMed] 40. Venko, K.; Roy Choudhury, A.; Novic, ˇ M. Computational Approaches for Revealing the Structure of Membrane Transporters: Case Study on Bilitranslocase. Comput. Struct. Biotech. J. 2017, 15, 232–242. [CrossRef] [PubMed] 41. Liang, J.; Naveed, H.; Jimenez-Morales, D.; Adamian, L.; Lin, M. Computational studies of membrane proteins: Models and predictions for biological understanding. Biochim. Biophys. Acta 2012, 1818, 927–941. [CrossRef] [PubMed] 42. Russel, D.; Lasker, K.; Webb, B.; Velázquez-Muriel, J.; Tjioe, E.; Schneidman-Duhovny, D.; Peterson, B.; Sali, A. Putting the pieces together: Integrative structure determination of macromolecular assemblies. PLoS Biol. 2012, 10, e1001244. [CrossRef] [PubMed] 43. Xu, Y.; Shen, Q.; Liu, X.; Lu, J.; Li, S.; Luo, C.; Gong, L.; Luo, X.; Zheng, M.; Jiang, H. Computational models for predicting interactions with membrane transporters. Curr. Med. Chem. 2013, 20, 2118–2136. [CrossRef] [PubMed] 44. Sedykh, A.; Fourches, D.; Duan, J.; Hucke, O.; Garneau, M.; Zhu, H.; Bonneau, P.; Tropsha, A. Human intestinal transporter database: QSAR modeling and virtual proﬁling of drug uptake, efﬂux and interactions. Pharm Res. 2013, 30, 996–1007. [CrossRef] [PubMed] 45. Chang, C.; Ray, A.; Swaan, P. In silico strategies for modeling membrane transporter function. Drug Discovery Today 2005, 10, 663–671. [CrossRef] 46. Mak, L.; Marcus, D.; Howlett, A.; Yarova, G.; Duchateau, G.; Klake, W.; Bender, A.; Glen, R.C. Metrabase: A cheminformatics and bioinformatics database for small molecule transporter data analysis and (Q)SAR modeling. J. Cheminform. 2015, 7, 31. [CrossRef] [PubMed] 47. Elbourne, L.D.H.; Tetu, S.; Hassan, K.; Paulsen, I.T. TransportDB 2.0: A database for exploring membrane transporters in sequenced genomes from all domains of life. Nucleic Acids Res. 2017, 45, D320–D324. [CrossRef] [PubMed] 48. Saier, M.H.; Reddy, V.S.; Tsu, B.V.; Ahmed, M.S.; Li, C.; Moreno-Hagelsieb, G. (The Transporter Classiﬁcation Database (TCDB): Recent advances. Nucleic Acids Res. 2016, 44, D372–D379. [CrossRef] [PubMed] 49. Battiston, L.; Passamonti, S.; Macagno, A.; Sottocasa, G.L. The bilirubin-binding motif of bilitranslocase and its relation to conserved motifs in ancient biliproteins. Biochem. Biophys. Res. Commun. 1998, 247, 687–692. [CrossRef] [PubMed] Molecules 2019, 24, 837 19 of 21 50. Niemi, M.; Pasanen, M.K.; Neuvonen, P.J. Organic anion transporting polypeptide 1B1: A genetically polymorphic transporter of major importance for hepatic drug uptake. Pharmacol. Rev. 2011, 63, 157–181. [CrossRef] [PubMed] 51. Passamonti, S.; Terdoslavich, M.; Margon, A.; Cocolo, A.; Medic, N.; Micali, F.; Decorti, G.; Franko, M. Uptake of bilirubin into HepG2 cells assayed by thermal lens spectroscopy. Function of bilitranslocase. FEBS J. 2005, 272, 5522–5535. [CrossRef] [PubMed] 52. Cvorovic, ´ J.; Passamonti, S. Membrane transporters for bilirubin and its conjugates: A systematic review. Front. Pharmacol. 2017, 8, 887. [CrossRef] [PubMed] 53. Karawajczyk, A.; Drgan, V.; Medic, N.; Oboh, G.; Passamonti, S.; Novic, ˇ M. Properties of ﬂavonoids inﬂuencing the binding to bilitranslocase investigated by neural network modelling. Biochem. Pharm. 2007, 73, 308–320. [CrossRef] [PubMed] 54. Župerl, Š.; Fornasaro, S.; Novic, ˇ M.; Passamonti, S. Experimental determination and prediction of bilitranslocase transport activity. Anal. Chim. Acta 2011, 705, 322–333. [CrossRef] [PubMed] 55. Maestro, A.; Terdoslavich, M.; Vanzo, A.; Kuku, A.; Tramer, F.; Nicolin, V.; Micali, F.; Decorti, G.; Passamonti, S. Expression of bilitranslocase in the vascular endothelium and its function as a ﬂavonoid transporter. Cardiovasc. Res. 2010, 85, 175–183. [CrossRef] [PubMed] 56. Nicolin, V.; Grill, V.; Micali, F.; Narducci, P.; Passamonti, S. Immunolocalisation of bilitranslocase in mucosecretory and parietal cells of the rat gastric mucosa. J. Mol. Histol. 2005, 36, 45–50. [CrossRef] [PubMed] 57. Brandoni, A.; Di Giusto, G.; Franca, R.; Passamonti, S.; Torres, A.M. Expression of kidney and liver bilitranslocase in response to acute biliary obstruction. Nephron. Physiol. 2010, 114, 35–40. [CrossRef] [PubMed] 58. Cvorovic, ´ J.; Ziberna, L.; Fornasaro, S.; Tramer, F.; Passamonti, S. Bioavailability of Flavonoids: The Role of Cell Membrane Transporters. In Polyphenols: Mechanisms of Action in Human Health and Disease, 2nd ed.; Watson, R.R., Preedy, V.R., Zibadi, S., Eds.; Elsevier Academic Press: Amsterdam, The Netherlands, 2018; pp. 295–320. 59. Martinci ˇ c, ˇ R.; Venko, K.; Župerl, Š.; Novic, ˇ M. Chemometrics approach for the prediction of structure-activity relationship for membrane transporter bilitranslocase. SAR QSAR Env. Res. 2014, 25, 853–872. [CrossRef] [PubMed] 60. Montanic, S.; Terdoslavich, M.; Rajcevic, U.; De Leo, L.; Bonin, S.; Curin Serbec, V.; Passamonti, S. Development and characterization of a novel mAb against bilitranslocase—A new biomarker of renal carcinoma. Radiol. Oncol. 2013, 47, 128–137. [CrossRef] [PubMed] 61. Golijanin, D.J.; Marshall, J.; Cardin, A.; Singer, E.A.; Wood, R.W.; Reeder, J.E.; Wu, G.; Yao, J.L.; Passamonti, S.; Messing, E.M. Bilitranslocase (BTL) is immunolocalised in proximal and distal renal tubules and absent in renal cortical tumors accurately corresponding to intraoperative near infrared ﬂuorescence (NIRF) expression of renal cortical tumors using intravenous indocyanine green (ICG). J. Urol. 2008, 179, 137. 62. Roy Choudhury, A.; Sikorska, E.; van den Boom, J.; Bayer, P.; Popenda, L.; Szutkowski, K.; Jurga, S.; Bonomi, M.; Sali, A.; Zhukov, I.; et al. Structural Model of the Bilitranslocase Transmembrane Domain Supported by NMR and FRET Data. PLoS ONE 2015, 10, e0135455. [CrossRef] [PubMed] 63. Passamonti, S.; Vrhovsek, U.; Vanzo, A.; Mattivi, F. The stomach as a site for anthocyanins absorption from food. FEBS Lett. 2003, 544, 210–213. [CrossRef] 64. Chirico, N.; Gramatica, P. Real External Predictivity of QSAR Models. Part 2. New Intercomparable Thresholds for Different Validation Criteria and the Need for Scatter Plot Inspection. J. Chem. Inf. Model. 2012, 52, 2044–2058. [CrossRef] [PubMed] 65. Todeschini, R.; Consonni, V.; Gramatica, P. Chemometrics in QSAR. In Comprehensive Chemometrics, Volume 4; Brown, S.D., Tauler, R., Walczak, B., Eds.; Elsevier: Oxford, UK, 2009; pp. 129–172. 66. Ruiz, I.L.; Gómez-Nieto, M.A. Study of the Applicability Domain of the QSAR Classiﬁcation Models by Means of the Rivality and Modelability Indexes. Molecules 2018, 23, 2756. [CrossRef] [PubMed] 67. Minovski, N.; Župerl, Š.; Drgan, V.; Novic, ˇ M. Assessment of applicability domain for multivariate counter-propagation artiﬁcial neural network predictive models by minimum Euclidean distance space analysis: A case study. Anal. Chim. Acta 2013, 759, 28–42. [CrossRef] [PubMed] 68. Gramatica, P.; Chirico, N.; Papa, E.; Kovarich, S.; Cassani, S. QSARINS: A New Software for the Development, Analysis, and Validation of QSAR MLR Models. J. Comput. Chem. 2013, 34, 2121–2132. [CrossRef] Molecules 2019, 24, 837 20 of 21 69. Marzo, M.; Kulkarni, S.; Manganaro, A.; Roncaglioni, A.; Wu, S.; Barton-Maclaren, T.S.; Lester, C.; Benfenati, E. Integrating in silico models to enhance predictivity for developmental toxicity. Toxicology 2016, 370, 127–137. [CrossRef] [PubMed] 70. Wang, X.; Wang, Q.; Morris, M.E. Pharmacokinetic interaction between the ﬂavonoid luteolin and gamma-hydoxybutyrate in rats: Potential involvement of monocarboxylate transporters. APPS J. 2008, 10, 47–55. 71. Piskula, M.K.; Yamakoshi, J.; Iwai, Y. Daizdein and genistein but not their glucosides are absorbed from the rat stomach. FEBS Lett. 1999, 447, 287–291. [CrossRef] 72. Roy, P.P.; Paul, S.; Mitra, I.; Roy, K. On two novel parameters for validation of predictive QSAR models. Molecules 2009, 14, 1660–1701. 73. Rodriguez-Morato, J.; Xicota, L.; Fito, M.; Farre, M.; Dierssen, M.; de la Torre, R. Potential role of olive oil phenolic compounds in the prevention of neurodegenerative diseases. Molecules 2015, 20, 4655–4680. [CrossRef] [PubMed] 74. Mathew, B.C.; Yoseph, B.A.; Dessale, T.; Daniel, R.S.; Alemayehu, A.; Campbell, I.W.; Augusti, K.T. Hypolipidaemic effect of leucodelphinidin derivative from Ficus bengalensis (Linn) on cholesterol fed rats. Res. J. Chem. Sci. 2012, 2, 54–60. 75. Terdoslavich, M.; de Graaf, I.A.M.; Proost, J.H.; Cocolo, A.; Passamonti, S.; Groothuis, G.M.M. Bilitranslocase is Involved in the Uptake of Bromosulfophthalein in Rat and Human Liver. Drug Metab. Lett. 2012, 6, 165–173. [CrossRef] [PubMed] 76. Bielen, L.; Kralj, I.; Curci ˇ c, ´ E.; Vodanovic, ´ M.; Boban, A.; Božina, N. Acute kidney injury, agranulocytosis, drug-induced liver injury, and posterior reversible encephalopathy syndrome caused by high-dose methotrexate—Possible role of low activity ABC and SLC drug transporters. Eu. J. Clinic. Pharmacol. 2018, 74, 1191–1192. [CrossRef] [PubMed] 77. Wilson, A.G.E. New Horizons in Predictive Drug Metabolism and Pharmacokinetics; Royal Society of Chemistry: Cambridge, UK, 2015; p. 422. 78. Zhang, G.; Ren, Y. Molecular Modeling and Design Studies of Purine Derivatives as Novel CDK2 Inhibitors. Molecules 2018, 23, 2924. [CrossRef] [PubMed] 79. Matyugina, E.; Khandazhinskaya, A.; Chernousova, L.; Andreevskaya, S.; Smirnova, T.; Chizhov, A.; Karpenko, I.; Kochetkov, S.; Alexandrova, L. The synthesis and antituberculosis activity of 50-nor carbocyclic uracil derivatives. Bioorg. Med. Chem. 2012, 20, 6680–6686. [CrossRef] [PubMed] 80. Okazaki, S.; Mizuhara, T.; Shimura, K.; Murayama, H.; Ohno, H.; Oishi, S.; Matsuoka, M.; Fujii, N. Identiﬁcation of anti-HIV agents with a novel benzo [4,5]isothiazolo[2,3-a]pyrimidine scaffold. Bioorg. Med. Chem. 2015, 23, 1447–1452. [CrossRef] [PubMed] 81. Sosnin, S.; Vashurina, M.; Withnall, M.; Karpov, P.; Fedorov, M.; Tetko, I.V. A Survey of Multi-Task Learning Methods in Chemoinformatics. Mol. Inf. 2018, 37, 1800108. [CrossRef] [PubMed] 82. Simões, R.S.; Maltarollo, V.G.; Oliveira, P.R.; Honorio, K.M. Transfer and Multi-task Learning in QSAR Modeling: Advances and Challenges. Front. Pharmacol. 2018, 9, 74. [CrossRef] [PubMed] 83. Žitnik, M.; Zupan, B. Matrix factorization-based data fusion for drug-induced liver injury prediction. Systems Biomed. 2014, 2, 16–22. [CrossRef] 84. Worth, A.P. The future of in silico chemical safety . . . and beyond. Comput. Toxicol. 2019, 10, 60–62. [CrossRef] 85. Dias, M.M.; Machado, N.F.L.; Marques, M.P.M. Dietary chromomes as antioxidant agents—the structural variable. Food Funct. 2011, 2, 595–602. [CrossRef] [PubMed] 86. Paiva-Martins, F.; Rodrigues, V.; Chalheiros, R.; Marques, M.P.M. Characterization of antioxidant olive oil biophenols by spectroscopic methods. J. Sci. Food Agric. 2011, 91, 309–314. [CrossRef] [PubMed] 87. Pisani, L.; Farina, R.; Nicolotti, O.; Gadaleta, D.; Soto-Otero, R.; Catto, M.; Di Braccio, M.; Mendez-Alvarez, E.; Carotti, A. In silico design of novel 2H-chromen-2-one derivatives as potent and selective MAO-B inhibitors. Eur. J. Med. Chem. 2015, 89, 98–105. [CrossRef] [PubMed] 88. Yildiz, S.Z.; Küçükislamoglu, ˘ M.; Tuna, M. Synthesis and characterization of novel ﬂavonoid-substituted phthalocyanines using ()naringenin. J. Organometal. Chem. 2009, 694, 4152–4161. [CrossRef] 89. Cao, G.; Soﬁc, E.; Prior, R.L. Antioxidant and prooxidant behavior of ﬂavonoids: Structure-activity relationships. Free Radical Biol. Med. 1997, 22, 749–760. [CrossRef] Molecules 2019, 24, 837 21 of 21 90. Petrussa, E.; Braidot, E.; Zancani, M.; Peresson, C.; Bertolini, A.; Patui, S.; Vianello, A. Plant Flavonoids—Biosynthesis, Transport and Involvement in Stress Responses. Int. J. Mol. Sci. 2013, 14, 14950–14973. [CrossRef] [PubMed] 91. Silva, F.A.M.; Borges, F.; Guimarães, C.; Lima, J.L.F.C.; Matos, C.; Reis, S. Phenolic acids and derivatives: Studies on the relationship among structure, radical scavenging activity, and physicochemical parameters. J. Agric. Food Chem. 2000, 48, 2122–2126. [CrossRef] [PubMed] 92. Heim, K.E.; Tagliaferro, A.R.; Bobilya, D.J. Flavonoid antioxidants: Chemistry, metabolism and structure-activity relationships. J. Nutr. Biochem. 2002, 13, 572–582. [CrossRef] 93. Martinci ˇ c, ˇ R.; Mravljak, J.; Švajger, U.; Perdih, A.; Andreluh, M.; Novic, ˇ M. In silico discovery of novel potent antioxidants on the basis of pulvinic acid and coumarin derivatives and their experimental evaluation. PLoS ONE 2015, 10, e0140602. [CrossRef] [PubMed] 94. Dewar, M.J.S.; Zoebisch, E.G.; Healy, E.F.; Stewart, J.J.P. Development and use of quantum mechanical molecular models. AM1: A new general purpose quantum mechanical molecular model. J. Am. Chem. Soc. 1985, 107, 3902–3909. 95. Katritzky, A.R.; Lobanov, V.S.; Karelson, M. Codessa 2.0. Comprehensive Descriptors for Structural and Statistical Analysis; University of Florida: Gainesville, FL, USA, 1994. 96. Dragon 7.0—Software for Molecular Descriptors Calculation. 2016. Available online: https://chm.kode-solutions. net/products_dragon.php (accessed on 20 August 2018). 97. Drgan, V.; Župerl, Š.; Vracko, ˇ M.; Cappelli, C.I.; Novic, ˇ M. CPANNatNIC software for counter-propagation neural network to assist in read-across. J. Cheminform. 2017, 9, 30. [CrossRef] [PubMed] 98. Drgan, V.; Župerl, Š.; Vracko, ˇ M.; Como, F.; Novic, ˇ M. Robust modelling of acute toxicity towards fathead minnow (Pimephales promelas) using counter-propagation artiﬁcial neural networks and genetic algorithm. SAR QSAR Environ. Res. 2016, 27, 501–519. [CrossRef] [PubMed] 99. Boughorbel, S.; Jarray, F.; El-Anbari, M. Optimal classiﬁer for imbalanced data using Matthews Correlation Coefﬁcient metric. PLoS ONE 2017, 12, e0177678. [CrossRef] [PubMed] 100. Fawcett, T. An introduction to ROC analysis. Pattern Recognition Lett. 2006, 27, 861–874. [CrossRef] 101. Metrabase. Available online: www-metrabase.ch.cam.ac.uk (accessed on 20 November 2018). 102. Chembench. Available online: https://chembench.mml.unc.edu (accessed on 4 December 2018). Models Availability: Newly developed classiﬁcation models for the prediction of bilitranslocase transport activity are available upon request from the authors. © 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/). http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Molecules Multidisciplinary Digital Publishing Institute http://www.deepdyve.com/lp/multidisciplinary-digital-publishing-institute/an-in-silico-approach-for-assessment-of-the-membrane-transporter-8NNlbaY5tE

Loading next page...

References (108)

A. Maestro, M. Terdoslavich, A. Vanzo, A. Kuku, Federica Tramer, V. Nicolin, F. Micali, G. Decorti, S. Passamonti (2010)
Expression of bilitranslocase in the vascular endothelium and its function as a flavonoid transporter.
Cardiovascular research, 85 1
R. Todeschini, Consonni, P. Gramatica (2009)
Chemometrics in QSAR
, 4
Hao Xie (2008)
Activity assay of membrane transport proteins.
Acta biochimica et biophysica Sinica, 40 4
Mariano Cárdenas, M. Marder, V. Blank, L. Roguin (2006)
Antitumor activity of some natural flavonoids and synthetic derivatives on various human and murine cancer cell lines.
Bioorganic & medicinal chemistry, 14 9
Rodolfo Simões, V. Maltarollo, P. Oliveira, K. Honório (2018)
Transfer and Multi-task Learning in QSAR Modeling: Advances and Challenges
Frontiers in Pharmacology, 9
O. Benavente-García, J. Castillo (2008)
Update on uses and properties of citrus flavonoids: new findings in anticancer, cardiovascular, and anti-inflammatory activity.
Journal of agricultural and food chemistry, 56 15
D. Golijanin, J. Marshall, A. Cardin, E. Singer, R. Wood, J. Reeder, Guan Wu, Jorge Yao, S. Passamonti, E. Messing (2008)
BILITRANSLOCASE (BTL) IS IMMUNOLOCALISED IN PROXIMAL AND DISTAL RENAL TUBULES AND ABSENT IN RENAL CORTICAL TUMORS ACCURATELY CORRESPONDING TO INTRAOPERATIVE NEAR INFRARED FLUORESCENCE (NIRF) EXPRESSION OF RENAL CORTICAL TUMORS USING INTRAVENOUS INDOCYANINE GREEN (ICG)
The Journal of Urology, 179
N. Minovski, S. Zuperl, V. Drgan, M. Novič (2013)
Assessment of applicability domain for multivariate counter-propagation artificial neural network predictive models by minimum euclidean distance space analysis: a case study.
Analytica chimica acta, 759
L. Pisani, R. Farina, O. Nicolotti, Domenico Gadaleta, R. Soto-Otero, M. Catto, M. Braccio, Estefanía Méndez-Álvarez, A. Carotti (2015)
In silico design of novel 2H-chromen-2-one derivatives as potent and selective MAO-B inhibitors.
European journal of medicinal chemistry, 89
R. Martinčič, K. Venko, S. Zuperl, M. Novič (2014)
Chemometrics approach for the prediction of structure–activity relationship for membrane transporter bilitranslocase£
SAR and QSAR in Environmental Research, 25
Shih‐Hua Fang, Y. Rao, Y. Tzeng (2005)
Inhibitory effects of flavonol glycosides from Cinnamomum osmophloeum on inflammatory mediators in LPS/IFN-gamma-activated murine macrophages.
Bioorganic & medicinal chemistry, 13 7
Francisco Silva, Fernanda Borges, C. Guimarães, José Lima, Carla Matos, Salette Reis (2000)
Phenolic acids and derivatives: studies on the relationship among structure, radical scavenging activity, and physicochemical parameters.
Journal of agricultural and food chemistry, 48 6
Pierre‐André Billat, Emilie Roger, S. Faure, F. Lagarce (2017)
Models for drug absorption from the small intestine: where are we and where are we going?
Drug discovery today, 22 5
(2012)
Putting the pieces together: Integrative structure determination of macromolecular assemblies
P. Gramatica, N. Chirico, E. Papa, S. Cassani, S. Kovarich (2013)
QSARINS: A new software for the development, analysis, and validation of QSAR MLR models
Journal of Computational Chemistry, 34
Egl≐ Sergedien≐, K. Jönsson, H. Szymusiak, B. Tyrakowska, I. Rietjens, N. C̆≐nas (1999)
Prooxidant toxicity of polyphenolic antioxidants to HL‐60 cells: description of quantitative structure‐activity relationships
FEBS Letters, 462
Andrew Worth (2019)
The future of in silico chemical safety … and beyond
Computational Toxicology
E. Matyugina, A. Khandazhinskaya, L. Chernousova, S. Andreevskaya, T. Smirnova, A. Chizhov, I. Karpenko, S. Kochetkov, L. Alexandrova (2012)
The synthesis and antituberculosis activity of 5'-nor carbocyclic uracil derivatives.
Bioorganic & medicinal chemistry, 20 22
L. Bielen, I. Kralj, Ela Ćurčić, M. Vodanović, A. Boban, Nada Božina (2018)
Acute kidney injury, agranulocytosis, drug-induced liver injury, and posterior reversible encephalopathy syndrome caused by high-dose methotrexate—possible role of low activity ABC and SLC drug transporters
European Journal of Clinical Pharmacology, 74
Kwangwoo Chun, Song-Kyu Park, Hwan‐Mook Kim, Yongseok Choi, Myung-Hwa Kim, Chun-Ho Park, Bo-Young Joe, T. Chun, Hyun-Moo Choi, Hee‐Yoon Lee, S. Hong, Myungbok Kim, Ky-Youb Nam, G. Han (2008)
Chromen-based TNF-α converting enzyme (TACE) inhibitors : Design, synthesis, and biological evaluation
Bioorganic & Medicinal Chemistry, 16
Jovana Čvorović, S. Passamonti (2017)
Membrane Transporters for Bilirubin and Its Conjugates: A Systematic Review
Frontiers in Pharmacology, 8
Z. Rasines-Perea, P. Teissèdre (2017)
Grape Polyphenols’ Effects in Human Cardiovascular Diseases and Diabetes
Molecules : A Journal of Synthetic Chemistry and Natural Product Chemistry, 22
FooDB version 1.0
A. Choudhury, E. Sikorska, J. Boom, P. Bayer, Łukasz Popenda, K. Szutkowski, S. Jurga, Massimiliano Bonomi, A. Sali, I. Zhukov, S. Passamonti, M. Novič (2015)
Structural Model of the Bilitranslocase Transmembrane Domain Supported by NMR and FRET Data
PLoS ONE, 10
J. Rodríguez-Morató, L. Xicota, M. Fitó, M. Farré, M. Dierssen, R. Torre (2015)
Potential Role of Olive Oil Phenolic Compounds in the Prevention of Neurodegenerative Diseases
Molecules, 20
Alan Wilson (2015)
New Horizons in Predictive Drug Metabolism and Pharmacokinetics
N. Chirico, P. Gramatica (2012)
Real External Predictivity of QSAR Models. Part 2. New Intercomparable Thresholds for Different Validation Criteria and the Need for Scatter Plot Inspection
Journal of chemical information and modeling, 52 8
Y. Rao, Shih‐Hua Fang, Y. Tzeng (2005)
Synthesis, growth inhibition, and cell cycle evaluations of novel flavonoid derivatives.
Bioorganic & medicinal chemistry, 13 24
R. Martinčič, Janez Mravljak, U. Švajger, A. Perdih, M. Anderluh, M. Novič (2015)
In Silico Discovery of Novel Potent Antioxidants on the Basis of Pulvinic Acid and Coumarine Derivatives and Their Experimental Evaluation
PLoS ONE, 10
Y. Xu, Q. Shen, X. Liu, J. Lu, S. Li, C. Luo, L. Gong, X. Luo, M. Zheng, H. Jiang (2013)
Computational models for predicting interactions with membrane transporters.
Current medicinal chemistry, 20 16
Teresa Serafim, Filipa Carvalho, M. Marques, R. Calheiros, T. Silva, J. Garrido, N. Milhazes, F. Borges, Fernanda Roleira, Elisiaario Silva, J. Holy, P. Oliveira (2011)
Lipophilic caffeic and ferulic acid derivatives presenting cytotoxicity against human breast cancer cells.
Chemical research in toxicology, 24 5
(2008)
Pharmacokinetic interaction between the flavonoid luteolin and gamma-hydoxybutyrate in rats: Potential involvement of monocarboxylate transporters
APPS J., 10
S. Yıldız, M. Kuçukislamoglu, Murat Tuna (2009)
Synthesis and characterization of novel flavonoid-substituted phthalocyanines using (±)naringenin
Journal of Organometallic Chemistry, 694
S. Passamonti, U. Vrhovšek, A. Vanzo, F. Mattivi (2003)
The stomach as a site for anthocyanins absorption from food 1
FEBS Letters, 544
P. Fresco, Fernanda Borges, Fernanda Borges, Carmen Diniz, M. Marques (2006)
New insights on the anticancer properties of dietary polyphenols
Medicinal Research Reviews, 26
M. Dias, N. Machado, M. Marques (2011)
Dietary chromones as antioxidant agents--the structural variable.
Food & function, 2 10
Ya-Ting Lee, M. Don, C. Liao, Hua-Wen Chiou, Chieh‐fu Chen, L. Ho (2005)
Effects of phenolic acid esters and amides on stimulus-induced reactive oxygen species production in human neutrophils.
Clinica chimica acta; international journal of clinical chemistry, 352 1-2
J. Dai, R. Mumper (2010)
Plant Phenolics: Extraction, Analysis and Their Antioxidant and Anticancer Properties
Molecules, 15
A. Bendini, L. Cerretani, A. Carrasco-Pancorbo, A. Gómez-Caravaca, A. Segura‐Carretero, A. Fernandez-Gutiérrez, G. Lercker (2007)
Phenolic molecules in virgin olive oils: a survey of their sensory properties, health effects, antioxidant activity and analytical methods. An overview of the last decade.
Molecules, 12 8
T. Nagaoka, A. Banskota, Y. Tezuka, I. Saiki, S. Kadota (2002)
Selective antiproliferative activity of caffeic acid phenethyl ester analogues on highly liver-metastatic murine colon 26-L5 carcinoma cell line.
Bioorganic & medicinal chemistry, 10 10
(1994)

Codessa 2.0. Comprehensive descriptors for structural and statistical analysis
L. Mennen, D. Sapinho, H. Ito, P. Galan, S. Hercberg, A. Scalbert (2008)
Urinary excretion of 13 dietary flavonoids and phenolic acids in free-living healthy subjects – variability and possible use as biomarkers of polyphenol intake
European Journal of Clinical Nutrition, 62
Metrabase
Jie Liang, Hammad Naveed, D. Jimenez-Morales, Larisa Adamian, Meishan Lin (2012)
Computational studies of membrane proteins: models and predictions for biological understanding.
Biochimica et biophysica acta, 1818 4
L. Mak, D. Marcus, Andrew Howlett, Galina Yarova, G. Duchateau, W. Klaffke, A. Bender, R. Glen (2015)
Metrabase: a cheminformatics and bioinformatics database for small molecule transporter data analysis and (Q)SAR modeling
Journal of Cheminformatics, 7
Aušra Nemeikaitė-Čėnienė, A. Imbrasaitė, E. Sergediene, N. Čėnas (2005)
Quantitative structure-activity relationships in prooxidant cytotoxicity of polyphenols: role of potential of phenoxyl radical/phenol redox couple.
Archives of biochemistry and biophysics, 441 2
(2016)

Dragon 7.0—Software for molecular descriptors calculation
M. Dewar, E. Zoebisch, Eamonn Healy, J. Stewart (1985)
Development and use of quantum mechanical molecular models. 76. AM1: a new general purpose quantum mechanical molecular model
Journal of the American Chemical Society, 107
(1980)
Plant phenolics. In Encyclopedia of Plant Physiology, Volume 8: Secondary Plant Products
Partha Roy, Somnath Paul, I. Mitra, K. Roy (2009)
On Two Novel Parameters for Validation of Predictive QSAR Models
Molecules, 14
V. Drgan, S. Zuperl, M. Vračko, F. Como, M. Novič (2016)
Robust modelling of acute toxicity towards fathead minnow (Pimephales promelas) using counter-propagation artificial neural networks and genetic algorithm
SAR and QSAR in Environmental Research, 27
A. Khoddami, Meredith Wilkes, T. Roberts (2013)
Techniques for Analysis of Plant Phenolic Compounds
Molecules, 18
(2018)
Bioavailability of Flavonoids: The Role of Cell Membrane Transporters
Fan Li, Jiajia Wu, Jin Wang, Xuelian Yang, Pei Cai, Qiaohong Liu, L. Kong, Xiaobing Wang (2017)
Synthesis and pharmacological evaluation of novel chromone derivatives as balanced multifunctional agents against Alzheimer's disease.
Bioorganic & medicinal chemistry, 25 14
(2006)
Antitumor activity of some natural flavonoids and derivatives on various human and maurine cancer cell lines
Catarina Gomes, Teresa Cruz, J. Andrade, N. Milhazes, F. Borges, M. Marques (2003)
Anticancer activity of phenolic acids of natural or synthetic origin: a structure-activity study.
Journal of medicinal chemistry, 46 25
G. Cao, E. Sofić, R. Prior (1997)
Antioxidant and prooxidant behavior of flavonoids: structure-activity relationships.
Free radical biology & medicine, 22 5
M. Piskuła, J. Yamakoshi, Yukihiko Iwai (1999)
Daidzein and genistein but not their glucosides are absorbed from the rat stomach
FEBS Letters, 447
Vermerris Wilfred, R. Nicholson (2006)
Phenolic Compound Biochemistry
L. Letenneur, C. Proust-Lima, A. Gouge, J. Dartigues, P. Barberger‐Gateau (2007)
Flavonoid intake and cognitive decline over a 10-year period.
American journal of epidemiology, 165 12
C. Manach, G. Williamson, C. Morand, A. Scalbert, C. Rémésy (2005)
Bioavailability and bioefficacy of polyphenols in humans. I. Review of 97 bioavailability studies.
The American journal of clinical nutrition, 81 1 Suppl
Liam Elbourne, S. Tetu, Karl Hassan, I. Paulsen (2016)
TransportDB 2.0: a database for exploring membrane transporters in sequenced genomes from all domains of life
Nucleic Acids Research, 45
V. Nicolin, V. Grill, F. Micali, P. Narducci, S. Passamonti (2005)
Immunolocalisation of bilitranslocase in mucosecretory and parietal cells of the rat gastric mucosa
Journal of Molecular Histology, 36
Y. Rao, Shih‐Hua Fang, Y. Tzeng (2004)
Differential effects of synthesized 2'-oxygenated chalcone derivatives: modulation of human cell cycle phase distribution.
Bioorganic & medicinal chemistry, 12 10
Daniel Russel, Keren Lasker, Ben Webb, J. Velázquez-Muriel, Elina Tjioe, D. Schneidman-Duhovny, Bret Peterson, A. Sali (2012)
Putting the Pieces Together: Integrative Modeling Platform Software for Structure Determination of Macromolecular Assemblies
PLoS Biology, 10
N. Machado, R. Calheiros, S. Fiuza, F. Borges, Alexandra Gaspar, J. Garrido, M. Marques (2007)
Phenolic esters with potential anticancer activity - the structural variable
Journal of Molecular Modeling, 13
Kelly Heim, A. Tagliaferro, D. Bobilya (2002)
Flavonoid antioxidants: chemistry, metabolism and structure-activity relationships.
The Journal of nutritional biochemistry, 13 10
Alexander Sedykh, D. Fourches, J. Duan, O. Hucke, M. Garneau, Hao Zhu, P. Bonneau, A. Tropsha (2013)
Human Intestinal Transporter Database: QSAR Modeling and Virtual Profiling of Drug Uptake, Efflux and Interactions
Pharmaceutical Research, 30
P. Pietta (2000)
Flavonoids as antioxidants.
Journal of natural products, 63 7
I. Fernandes, Rosa Pérez-Gregorio, Susana Soares, N. Mateus, V. Freitas (2017)
Wine Flavonoids in Health and Disease Prevention
Molecules : A Journal of Synthetic Chemistry and Natural Product Chemistry, 22
S. Okazaki, T. Mizuhara, Kazuya Shimura, Hiroto Murayama, H. Ohno, S. Oishi, M. Matsuoka, N. Fujii (2015)
Identification of anti-HIV agents with a novel benzo[4,5]isothiazolo[2,3-a]pyrimidine scaffold.
Bioorganic & medicinal chemistry, 23 7
S. Montanic, M. Terdoslavich, U. Rajčevič, Luigina Leo, S. Bonin, V. Šerbec, S. Passamonti (2013)
Development and characterization of a novel mAb against bilitranslocase - a new biomarker of renal carcinoma
Radiology and Oncology, 47
P. Thompson (1965)
Plant Phenolics
Nature, 207
Sabri Boughorbel, Fethi Jarray, Mohammed El-Anbari (2017)
Optimal classifier for imbalanced data using Matthews Correlation Coefficient metric
PLoS ONE, 12
Gaomin Zhang, Y. Ren (2018)
Molecular Modeling and Design Studies of Purine Derivatives as Novel CDK2 Inhibitors
Molecules, 23
I. Ruiz, M. Gómez-Nieto (2018)
Study of the Applicability Domain of the QSAR Classification Models by Means of the Rivality and Modelability Indexes
Molecules, 23
M. Saier, V. Reddy, Brian Tsu, Muhammad Ph.D., Chun Li, G. Moreno-Hagelsieb (2015)
The Transporter Classification Database (TCDB): recent advances
Nucleic Acids Research, 44
L. Battiston, S. Passamonti, A. Macagno, G. Sottocasa (1998)
The bilirubin-binding motif of bilitranslocase and its relation to conserved motifs in ancient biliproteins.
Biochemical and biophysical research communications, 247 3
D. Rio, A. Rodriguez-Mateos, J. Spencer, M. Tognolini, G. Borges, A. Crozier (2013)
Dietary (poly)phenolics in human health: structures, bioavailability, and evidence of protective effects against chronic diseases.
Antioxidants & redox signaling, 18 14
E. Petrussa, E. Braidot, M. Zancani, C. Peresson, A. Bertolini, Sonia Patui, A. Vianello (2013)
Plant Flavonoids—Biosynthesis, Transport and Involvement in Stress Responses
International Journal of Molecular Sciences, 14
T. Dessale, A. Alemayehu (2012)
Hypolipidaemic effect of leucodelphinidin derivative from Ficus bengalensis Linn on cholesterol fed rats
M. Marzo, S. Kulkarni, A. Manganaro, A. Roncaglioni, Shengde Wu, T. Barton-Maclaren, C. Lester, E. Benfenati (2016)
Integrating in silico models to enhance predictivity for developmental toxicity.
Toxicology, 370
V. Blank, C. Poli, M. Marder, L. Roguin (2004)
Antiproliferative activity of various flavonoids and related compounds: additive effect of interferon-alpha2b.
Bioorganic & medicinal chemistry letters, 14 1
Zhongbing Lu, Guangjun Nie, P. Belton, Huiru Tang, Baolu Zhao (2006)
Structure–activity relationship analysis of antioxidant ability and neuroprotective effect of gallic acid derivatives
Neurochemistry International, 48
Xiaodong Wang, Qi Wang, M. Morris (2008)
Pharmacokinetic Interaction between the Flavonoid Luteolin and γ-Hydroxybutyrate in Rats: Potential Involvement of Monocarboxylate Transporters
The AAPS Journal, 10
Cheng Chang, A. Ray, P. Swaan (2005)
In silico strategies for modeling membrane transporter function.
Drug discovery today, 10 9
Chembench
M. Niemi, M. Pasanen, P. Neuvonen (2011)
Organic Anion Transporting Polypeptide 1B1: a Genetically Polymorphic Transporter of Major Importance for Hepatic Drug Uptake
Pharmacological Reviews, 63
F. Paiva‐Martins, Vera Rodrigues, R. Calheiros, M. Marques (2011)
Characterization of antioxidant olive oil biophenols by spectroscopic methods.
Journal of the science of food and agriculture, 91 2
A. Karawajczyk, V. Drgan, Nevenka Medic, G. Oboh, S. Passamonti, M. Novič (2007)
Properties of flavonoids influencing the binding to bilitranslocase investigated by neural network modelling.
Biochemical pharmacology, 73 2
M. Terdoslavich, I. Graaf, J. Proost, Alessandra Cocolo, S. Passamonti, G. Groothuis (2013)
Bilitranslocase is involved in the uptake of bromosulfophthalein in rat and human liver.
Drug metabolism letters, 6 3
M. Khalil, S. Sulaiman (2010)
The potential role of honey and its polyphenols in preventing heart diseases: a review.
African journal of traditional, complementary, and alternative medicines : AJTCAM, 7 4
Claudriana Locatelli, Rober Rosso, M. Santos-Silva, C. Souza, M. Licínio, P. Leal, M. Bazzo, R. Yunes, T. Creczynski‐Pasa (2008)
Ester derivatives of gallic acid with potential toxicity toward L1210 leukemia cells.
Bioorganic & medicinal chemistry, 16 7
S. Fiuza, Catarina Gomes, L. Teixeira, M. Cruz, M. Cordeiro, N. Milhazes, Fernanda Borges, M. Marques (2004)
Phenolic acid derivatives with potential anticancer properties--a structure-activity relationship study. Part 1: methyl, propyl and octyl esters of caffeic and gallic acids.
Bioorganic & medicinal chemistry, 12 13
M. Zitnik, B. Zupan (2014)
Matrix factorization-based data fusion for drug-induced liver injury prediction
Systems Biomedicine, 2
(1994)
Comprehensive Descriptors for Structural and Statistical Analysis; University of Florida
A. Brandoni, Gisela Giusto, R. Franca, S. Passamonti, A. Torres (2010)
Expression of Kidney and Liver Bilitranslocase in Response to Acute Biliary Obstruction
Nephron Physiology, 114
V. Drgan, S. Zuperl, M. Vračko, C. Cappelli, M. Novič (2017)
CPANNatNIC software for counter-propagation neural network to assist in read-across
Journal of Cheminformatics, 9
S. Passamonti, M. Terdoslavich, R. Franca, A. Vanzo, Federica Tramer, E. Braidot, E. Petrussa, A. Vianello (2009)
Bioavailability of flavonoids: a review of their membrane transport and the function of bilitranslocase in animal and plant organisms.
Current drug metabolism, 10 4
Dylan Marsh, Sukanya Das, J. Ridell, S. Smid (2017)
Structure-activity relationships for flavone interactions with amyloid β reveal a novel anti-aggregatory and neuroprotective effect of 2',3',4'-trihydroxyflavone (2-D08).
Bioorganic & medicinal chemistry, 25 14
S. Sosnin, M. Vashurina, M. Withnall, Pavel Karpov, M. Fedorov, I. Tetko (2018)
A Survey of Multi‐task Learning Methods in Chemoinformatics
Molecular Informatics, 38
B. Bauvois, Marie-Line Puiffe, J. Bongui, S. Paillat, C. Monneret, D. Dauzonne* (2003)
Synthesis and biological evaluation of novel flavone-8-acetic acid derivatives as reversible inhibitors of aminopeptidase N/CD13.
Journal of medicinal chemistry, 46 18
S. Zuperl, Stefano Fornasaro, M. Novič, S. Passamonti (2011)
Experimental determination and prediction of bilitranslocase transport activity.
Analytica chimica acta, 705 1-2
K. Venko, A. Choudhury, M. Novič (2017)
Computational Approaches for Revealing the Structure of Membrane Transporters: Case Study on Bilitranslocase
Computational and Structural Biotechnology Journal, 15
S. Passamonti, M. Terdoslavich, A. Margon, Alessandra Cocolo, Nevenka Medic, F. Micali, G. Decorti, M. Franko (2005)
Uptake of bilirubin into HepG2 cells assayed by thermal lens spectroscopy
The FEBS Journal, 272
Tom Fawcett (2006)
An introduction to ROC analysis
Pattern Recognit. Lett., 27
(2018)
Bioavailability of Flavonoids: The Role of Cell Membrane Transporters. In Polyphenols: Mechanisms of Action in Human Health and Disease
T. Bruyn, G. Westen, A. IJzerman, B. Stieger, P. Witte, P. Augustijns, P. Annaert (2013)
Structure-Based Identification of OATP1B1/3 Inhibitors
Molecular Pharmacology, 83

Publisher: Multidisciplinary Digital Publishing Institute
Copyright: © 1996-2019 MDPI (Basel, Switzerland) unless otherwise stated
ISSN: 1420-3049
DOI: 10.3390/molecules24050837
Publisher site: See Article on Publisher Site

Abstract

molecules Article An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase Katja Venko * and Marjana Novic ˇ * Laboratory for Cheminformatics, Theory Department, National Institute of Chemistry, SI-1000 Ljubljana, Slovenia * Correspondence: [email protected] (K.V.); [email protected] (M.N.); Tel.: +386-1-4760253 (M.N.) Received: 1 February 2019; Accepted: 26 February 2019; Published: 27 February 2019 Abstract: Phenols are the most abundant naturally accessible antioxidants present in a human normal diet. Since numerous beneﬁcial applications of phenols as preventive agents in various diseases were revealed, the evaluation of phenols bioavailability is of high interest of researchers, consumers and drug manufacturers. The hydrophilic nature of phenols makes a cell membrane penetration difﬁcult, which imply an alternative way of uptake via membrane transporters. However, the structural and functional data of membrane transporters are limited, thus the in silico modelling is really challenging and urgent tool in elucidation of transporter ligands. Focus of this research was a particular transporter bilitranslocase (BTL). BTL has a broad tissue expression (vascular endothelium, absorptive and excretory epithelia) and can transport wide variety of poly-aromatic compounds. With available BTL data (pKi [mmol/L] for 120 organic compounds) a robust and reliable QSAR models for BTL transport activity were developed and extrapolated on 300 phenolic compounds. For all compounds the transporter proﬁles were assessed and results show that dietary phenols and some drug candidates are likely to interact with BTL. Moreover, synopsis of predictions from BTL models and hits/predictions of 20 transporters from Metrabase and Chembench platforms were revealed. With such joint transporter analyses a new insights for elucidation of BTL functional role were acquired. Regarding limitation of models for virtual proﬁling of transporter interactions the computational approach reported in this study could be applied for further development of reliable in silico models for any transporter, if in vitro experimental data are available. Keywords: phenols; membrane transporters; bilitranslocase; transport activity; QSAR; in silico models 1. Introduction Phenols are one of the most promising groups of dietary preventive agents [1,2]. They are widely amenable natural and nutritional agents, i.e. nutraceuticals, which belong to a heterogeneous family of chemical compounds comprising benzoquinones, phenolic acids and aldehydes, coumarins, chromones, xanthonoids, stilbenoids, ﬂavonoids, lignans, tannins, etc. [3]. In general, phenolic compounds are classiﬁed based on the number of phenol units in the molecule, as simple phenols or polyphenols [2,4]. Functionally, they are secondary metabolites produced by plants and microorganisms or synthesized industrially [5,6]. As the most abundant antioxidants present in a human normal diet they have received increasing interest from researchers, consumers, food and drug manufacturers [2,7]. The phenols, being recognized as one of the most potential antioxidants, have role in preventing an imbalance in reactive oxygen species (ROS) production and an attenuation Molecules 2019, 24, 837; doi:10.3390/molecules24050837 www.mdpi.com/journal/molecules Molecules 2019, 24, 837 2 of 21 of anti-oxidant defense, what can turnout the genesis of many ageing phenomena and diseases [1,8–10]. Furthermore, bioactivities of phenols appeared to be quite broad-spanning, yet many multi-step interactions have not been exactly elucidated. Thus, numerous beneﬁcial application of phenols as preventive and therapeutic agents in cancer and inﬂammatory, neurodegenerative and cardiovascular diseases have been reported [1,9,11–32]. Particularly, the initiation stages of carcinogenesis were shown to be largely preventable and susceptible to modulation by nutraceuticals, especially highlighting the dietary polyphenols as one of the most promising anti-carcinogenic agents [10,15]. However, solely the normal diet phenolic intake is insufﬁcient for the effective chemoprevention, but the ingestion of extra dietary phenolic supplements as support to chemotherapy is a promising approach [7,10]. The epidemiological data show that life-long intake of phenolic compounds is beneﬁcial and related to reduce the incidence of various diseases [8,11,15,33], yet to improve our knowledge on the bioavailability of polyphenols from the diet uptake is of high interest. The uptake of nutraceuticals and orally administrated drugs is dependent on their capacity of absorption from the digestive tract. The studies, which followed the oral ingestion and gastro-intestinal uptake of ﬂavonoids, show that the concentrations are <1% of the administrated dose [34,35]. Such stringent intestinal control can be explained with the experimental evidence that many importers are unable to transport phenols, while various exporters can [34]. This complexity obscures the delivery of phenolic nutraceuticals or drugs, and raises the need for good substrate permeability predictive models. Since phenols are widely distributed in fruits, vegetables, grains and their processed foodstuffs, but are not synthesized by animal cells, their detection in animals is solely indicative of plant ingestion [36]. Therefore, their absorption and tissue distribution in specimens is indispensably dependent on transport ability of membrane transporters, and in this regard the cellular uptake of phenolic compounds can be efﬁciently estimated by in vitro transport assays [37–39], while in silico applications of membrane transport processes are very limited due to extremely slow rise of high-resolution structures of membrane transporters [40]. At present, the development of in silico methodologies for ﬁlling the gap among protein structures and properties is emerging and helps circumventing the difﬁculties in resolving membrane transporter structures and functions [40–42]. In general, two in silico strategies exists: (i) ligand-based methods enabling analysis of transporter substrate speciﬁcity in the absence of 3D structure, and (ii) protein-based methods providing modeling of transporters structures based on sequence identity of available experimentally determined atomic structures [43]. Among ﬁrst, the pharmacophore mapping and development of quantitative structure-activity relationship (QSAR) models are the most common for predicting ligand binding modes (Figure 1) [44]. Among second, the comparative modeling is the most frequently used approach to reveal structure and function of transporter, yet is only applicable, if homolog structures are available [45]. Furthermore, studies show that the combination of both strategies is favorable as it provides information beyond individual models [45]. Due to shortage of currently available experimental functional and structural data on membrane transporters, the in silico modelling is essential in the discovery of adverse effects, ADMET properties and more efﬁcient transporter substrates (e.g., dietary products, drugs) [44]. Molecules 2019, 24, x FOR PEER REVIEW 3 of 21 transporters studies. Nevertheless, according to the reports in TransportDB (www.membranetransport.org) [47] and Transporter Classification Database (www.tcdb.org) [48] a hundreds of potential transporters exist (in the human genome > 1400 unique transport proteins were identified). Therefore, our interest arises on a specific, neglected but extremely attractive Molecules 2019, 24, 837 3 of 21 transporter bilitranslocase (BTL, Uniprot O88750, GenBank: Y12178.1). Figure Figure 1. 1. General scheme o General scheme of f the quantitative structur the quantitative structure-activity relatio e-activity relationsh nship (QSAR) ip (QSAR) strategy. strategy. Chemical structures and biological activity data are obligated input for model building and Chemical structures and biological activity data are obligated input for model building and validation. After validat the ion. model After the mode development l de and velopment an optimization d optimizat the new ion activity the new act predictions ivitycan predicti be made ons c based an be made solely on chemical structure input. based solely on chemical structure input. By evidence dietary substances and drugs in mammals are transported via membrane Bilitranslocase is a 38.22 kDa bilirubin membrane transporter and is the first reported member transporters rather than passive diffusion [38]. Therefore, it is of interest to proﬁle the membrane of the Bilirubin Transporter (BRT) Family (TCDB 2.A.65.1.1) [35]. The gene of BTL is unidentified, transport ability by using the recently available transporter databases in the public domain like and even no homology with any other known protein structure was found [35,49]. Yet, a surprising Metrabase (www-metrabase.ch.cam.ac.uk) [46] and ChemBench (www.chembench.mml.unc.edu) [44]. feature of BTL is its coding sequence, which is 94% homologous to a segment of the antisense strand Metrabase provides the structural, physiochemical and biological data about transporters and their of the ceruloplasmin transcript (RNO 30106) in the anti-CODE database (www.anticode.org). The ligands. It contains 16 efﬂux and inﬂux transporters, which are also included in the guidance for considerable biological importance in BTL sequence has the bilirubin-binding motif, which plays a the approval of drugs proposed by the US Food and Drug Administration and by the European central role in ligand binding, and it is similar to highly conserved segment in α-phycocyanins [49], Medicines Agency [46]. On the other hand, Chembench platform offers the largest collection of which are ancient biliproteins present in cyanobacteria. Further on, BTL shares similar tissue human intestinal transporters interactions and the compendium of QSAR models for virtual screening distribution and functional properties with some of the members of the family of organic anion of chemicals. Moreover, these two databases reﬂect the current research focus in mammalian transporters (OAT) (Figure 2), which is the most dominant family of the solute carriers (SLC, one of transporters, as up to the best of our knowledge they encompass majority of existing transporters the two classified categories of transporters) [50]. BTL does not transport only bilirubin into studies. Nevertheless, according to the reports in TransportDB (www.membranetransport.org) [47] hepatocytes [51,52], but can transport a wide variety of poly-aromatic compounds among vitally and Transporter Classiﬁcation Database (www.tcdb.org) [48] a hundreds of potential transporters exist important endogenous substrates and drugs [35,53,54]. Its tissue expression is broad from vascular (in the human genome > 1400 unique transport proteins were identiﬁed). Therefore, our interest arises endothelium to stomach, intestine, kidney and liver epithelia (Figure 2) [55–58]. These peculiarities on a speciﬁc, neglected but extremely attractive transporter bilitranslocase (BTL, Uniprot O88750, make BTL a hot spot for future investigation as potential important endogenous drug delivery GenBank: Y12178.1). system to the intracellular targets, especially for its ability of cancer drugs uptake [59] and also as Bilitranslocase is a 38.22 kDa bilirubin membrane transporter and is the ﬁrst reported member tumor biomarker [60]. The study on renal carcinoma cells figure out an outstanding fact, that BTL of the Bilirubin Transporter (BRT) Family (TCDB 2.A.65.1.1) [35]. The gene of BTL is unidentiﬁed, expression is severely down-regulated in the renal cortical tumors [61], therefore the BTL assessment and even no homology with any other known protein structure was found [35,49]. Yet, a surprising with specific BTL antibodies could serve as a novel biomarker of human kidney cancers [60]. In spite feature of BTL is its coding sequence, which is 94% homologous to a segment of the antisense of our high interest and several efforts, the synthesis of BTL was unsuccessful, thus its structural and strand of the ceruloplasmin transcript (RNO 30106) in the anti-CODE database (www.anticode.org). functional characterization remains unclear. However, up to date the data on tissue expression The considerable biological importance in BTL sequence has the bilirubin-binding motif, which plays based on the specific BTL monoclonal antibodies [35], transport activity assays [53,54] and a central role in ligand binding, and it is similar to highly conserved segment in -phycocyanins [49], which are ancient biliproteins present in cyanobacteria. Further on, BTL shares similar tissue distribution and functional properties with some of the members of the family of organic anion transporters (OAT) (Figure 2), which is the most dominant family of the solute carriers (SLC, Molecules 2019, 24, 837 4 of 21 one of the two classiﬁed categories of transporters) [50]. BTL does not transport only bilirubin into hepatocytes [51,52], but can transport a wide variety of poly-aromatic compounds among vitally important endogenous substrates and drugs [35,53,54]. Its tissue expression is broad from vascular endothelium to stomach, intestine, kidney and liver epithelia (Figure 2) [55–58]. These peculiarities make BTL a hot spot for future investigation as potential important endogenous drug delivery system to the intracellular targets, especially for its ability of cancer drugs uptake [59] and also as tumor biomarker [60]. The study on renal carcinoma cells ﬁgure out an outstanding fact, that BTL expression is severely down-regulated in the renal cortical tumors [61], therefore the BTL assessment with speciﬁc BTL antibodies could serve as a novel biomarker of human kidney cancers [60]. In spite of our high interest and several efforts, the synthesis of BTL was unsuccessful, thus its structural and functional characterization remains unclear. However, up to date the data on tissue Molecules 2019, 24, x FOR PEER REVIEW 4 of 21 expression based on the speciﬁc BTL monoclonal antibodies [35], transport activity assays [53,54] and computational elucidation of its 3D structure based on available NMR data (Figure 2) [62] are computational elucidation of its 3D structure based on available NMR data (Figure 2) [62] are available. available Since BTL . Sin gene ce BTL is not g deﬁned, ene is not the de BTL fincharacterization ed, the BTL char experimentally acterization exp still eri rm emains entally limited still re only mains on limited only on immunological approach by specific ant-peptide antibodies and in silico modeling immunological approach by speciﬁc ant-peptide antibodies and in silico modeling with identiﬁed wi synthetic th identif transmembrane ied synthetic tra segments. nsmembrane segments. Figure 2. (a) Structural model of membrane transporter bilitranslocase (top-scoring computational Figure 2. (a) Structural model of membrane transporter bilitranslocase (top-scoring computational arrangement of four transmembrane helices based on NMR spectroscopy data) [62]. No homologs arrangement of four transmembrane helices based on NMR spectroscopy data) [62]. No homologs were were defined for 1026 bp mRNA sequence (GenBank: Y12178.1), thus the comparative modeling deﬁned for 1026 bp mRNA sequence (GenBank: Y12178.1), thus the comparative modeling studies studies are not possible to perform. The further structure characterization of bilitranslocase is under are not possible to perform. The further structure characterization of bilitranslocase is under study. study. (b) Scheme of cell membrane transporters distribution in different tissues [44,50,58]. ABC (b) Scheme of cell membrane transporters distribution in different tissues [44,50,58]. ABC transporters transporters ar are indicated e indicate by black arr d by ows black arrows a and solute carriers nd solute (SLC) carriers (SLC) by w by white arrows; hite arrows; on one direction e direction transport transport (one (one way arrow), way arrow), bidirectional tran bidirectional transport (two sport (two way way arrow). arrow). Considering the current limitation and dearth of empirical data about the interesting BTL transporter, the usage of in silico approaches is more than welcome to elucidate the BTL functional properties. Moreover, they can even provide new guidelines for further design of the BTL experiments. Some QSAR models for prediction of the BTL transport ability were preliminary build on available data of biological assay of the BTL transport activity for the poly-aromatic organic compounds [53,54,59]. The in vitro experiments showed fast kinetics of plasma appearance of polyphenols and their concentrations were found in the range of half-saturation of BTL in the portal blood [63]. Thus, Passamonti et al. suggest that particularly at the gastric level the BTL-mediated mechanism could be the basis of the fast kinetics and ability of polyphenols to permeate the gastric mucosa [35,63]. Even more, it seems that the fast absorption of polyphenols in the upper digestive tract is a mindful steep, as polyphenols in the lower segments of digestive tract are exposed to intensive chemical and microbial degradation [1,34], which significantly affect their bioavailability. The aim of this study was to assess the membrane transporters activity of selected phenolic compounds by a comprehensive virtual profiling. Firstly, we developed new robust and reliable models for BTL transport activity that were used for screening of new untested dataset of 300 Molecules 2019, 24, 837 5 of 21 Considering the current limitation and dearth of empirical data about the interesting BTL transporter, the usage of in silico approaches is more than welcome to elucidate the BTL functional properties. Moreover, they can even provide new guidelines for further design of the BTL experiments. Some QSAR models for prediction of the BTL transport ability were preliminary build on available data of biological assay of the BTL transport activity for the poly-aromatic organic compounds [53,54,59]. The in vitro experiments showed fast kinetics of plasma appearance of polyphenols and their concentrations were found in the range of half-saturation of BTL in the portal blood [63]. Thus, Passamonti et al. suggest that particularly at the gastric level the BTL-mediated mechanism could be the basis of the fast kinetics and ability of polyphenols to permeate the gastric mucosa [35,63]. Even more, it seems that the fast absorption of polyphenols in the upper digestive tract is a mindful steep, as polyphenols in the lower segments of digestive tract are exposed to intensive chemical and microbial degradation [1,34], which signiﬁcantly affect their bioavailability. The aim of this study was to assess the membrane transporters activity of selected phenolic compounds by a comprehensive virtual proﬁling. Firstly, we developed new robust and reliable models for BTL transport activity that were used for screening of new untested dataset of 300 phenolic compounds (including potential drug candidates, nutraceuticals). With the developed QSAR models the new insights into transport interactions of structurally unresolved BTL transmembrane protein were gained and the phenolic compounds that could inhibit the BTL transport were identiﬁed. We also discussed the preferences of using the consensus of models to enhance predictivity of in silico models. Secondly, the unique transporter proﬁle of studied compounds was performed based on the results from a new BTL models developed in this study and from the results obtained from Metrabase [46] and Chembench platforms [44], a freely available tools for membrane transporters characterization. 2. Results 2.1. BTL Classiﬁcation Models The development of QSAR models is a complex process therefore general OECD principles to properly produce and validate models [64,65] have been taken into account. In order to build reliable models and also to detect possible model pathologies in the modelling strategy, a cross validation and external validation was included. Thus, BTL dataset (n = 120) was divided into two or three subsets in rate 75/25 or 60/25/15. The model NN-C had three subsets; training set (n = 70), test set (n = 31) and validation set (n = 19). Models NN-D and Q-D had two subsets; training set (n = 90) and validation set (n = 30). A dataset splitting conditions are precisely stated in study of Martinci ˇ c ˇ et al. [59]. Initial modeling datasets included 66 or 78 variables, Codessa and Dragon descriptors, respectively. The model NN-C was the best model available from study of Martinci ˇ c ˇ et al. [59] and was developed with non-reduced number of descriptors (66 Codessa MDs). In this study new Dragon molecular descriptors (MDs) were calculated and further model optimization with cross validation and genetic algorithm was used. The newly developed models (NN-D and Q-D) contain signiﬁcantly reduced set of MDs (from 78 to 18/11). The list of selected descriptors of NN-D and Q-D models is represented in Table S4 (Supplementary Material). The selected models have comparable quality parameters for training set, yet new CP-ANN model has signiﬁcantly improved performance of validation set (Tables 1 and 2). Regarding results of quantitative quality indicators and graphical quality parameter (ROC curve) the NN-D model shows the best training and validation performances (Figure 3). Predictions for compounds used in the models development and validation are presented in Table S1 (Supplementary Material). Molecules 2019, 24, x FOR PEER REVIEW 5 of 21 phenolic compounds (including potential drug candidates, nutraceuticals). With the developed QSAR models the new insights into transport interactions of structurally unresolved BTL transmembrane protein were gained and the phenolic compounds that could inhibit the BTL transport were identified. We also discussed the preferences of using the consensus of models to Molecules 2019, 24, 837 6 of 21 enhance predictivity of in silico models. Secondly, the unique transporter profile of studied compounds was performed based on the results from a new BTL models developed in this study Table 1. Statistic parameters of the best three single models and consensus classiﬁcation models. and from the results obtained from Metrabase [46] and Chembench platforms [44], a freely available tools for membrane transporters characterization. Single Models 2-Model Consensus 3-Model Consensus NN-C + NN-C + NN-C + 2. Results NN-C NN-D Q-D A B A + B NN-D Q-D Q-D TP 44 44 46 33 33 41 29 19 48 2.1. BTL Classification Models TN 61 65 58 47 49 55 45 17 62 FP 9 5 12 6 2 1 1 7 8 The development of QSAR models is a complex process therefore general OECD principles to FN 6 6 4 0 2 0 0 2 2 properly produce and validate models [64,65] have been taken into account. In order to build 120 120 120 86 86 97 75 45 120 reliable models and also to detect possible model pathologies in the modelling strategy, a cross PR 1 (0.98) 1 (0.92) 1 (0.97) 0.72 0.72 0.81 0.62 0.37 1 validation and external validation was included. Thus, BTL dataset (n = 120) was divided into two or SE 0.88 0.88 0.92 1.00 0.94 1.00 1.00 0.90 0.96 three subsets SP in ra 0.87 te 75/25 0.93 or 60/25/0.83 15. The mo 0.89 del NN-0.96 C had three 0.98 subsets; tra 0.98 ini 0.71 ng set ( 0.89 n = 70), test ACC 0.88 0.91 0.87 0.93 0.95 0.99 0.99 0.80 0.92 set (n = 31) and validation set (n = 19). Models NN-D and Q-D had two subsets; training set (n = 90) NPV 0.91 0.92 0.94 1.00 0.96 1.00 1.00 0.89 0.97 and validation set (n = 30). A dataset splitting conditions are precisely stated in study of Martinčič et PPV 0.83 0.90 0.79 0.85 0.94 0.98 0.97 0.73 0.86 al. [59]. MCC 0.75 0.81 0.74 0.87 0.90 0.98 0.97 0.62 0.84 Initial modeling datasets included 66 or 78 variables, Codessa and Dragon descriptors, A—complete agreement among predictions, B—maximal one disagreement among predictions. respectively. The model NN-C was the best model available from study of Martinčič et al. [59] and was developed with non-reduced number of descriptors (66 Codessa MDs). In this study new Table 2. Summary of classiﬁcation models results. Dragon molecular descriptors (MDs) were calculated and further model optimization with cross No. of Active validation and genetic algorithm was used. The newly develope Predict d Ratio models (NN-D an [%] d Q-D) contain No. of Compound Predictions Type ACC ACC Model ID. TR V Descriptors significantly reduced set of MDs (from 78 to 18/11). The list of selected descriptors of NN-D and Q-D BTL Dataset Phenols BTL Dataset Phenols models is represented in Table S4 (Supplementary Material). The selected models have comparable NN-C CP-ANN 66 0.93 0.74 100 (97) 100 (94) 54 (51) 75 (74) quality parameters for training set, yet new CP-ANN model has significantly improved performance NN-D CP-ANN 18 0.92 0.87 100 (92) 100 (69) 49 (46) 138 (100) Q-D MLR 11 0.88 0.77 100 (97) 100 (93) 58 (55) 72 (68) of validation set (Tables 1 and 2). Regarding results of quantitative quality indicators and graphical A + B 3 cons / 0.93 0.87 100 (99) 100 (100) 56 (55) 65 (65) quality parameter (ROC curve) the NN-D model shows the best training and validation A 3 cons / 1.00 0.92 62 (62) 36 (36) 30 (30) 15 (15) performances (Figure 3). Predictions for compounds used in the models development and validation NN-D + Q-D 2 cons / 1.00 0.96 81 (81) 50 (50) 42 (42) 31 (31) are presented in T ( ) values ablefor S1 compounds (Supplement in AD; ary M 3 cons—3-model aterial). consensus; 2 cons—2-model consensus. Figure 3. ROC curves of the three selected classiﬁcation models: (a) training set, (b) validation set. Figure 3. ROC curves of the three selected classification models: (a) training set, (b) validation set. The reliability of a QSAR model depends on its capacity to achieve conﬁdent predictions of new The reliability of a QSAR model depends on its capacity to achieve confident predictions of new compounds not considered in the model development, which implies the necessity of considering compounds not considered in the model development, which implies the necessity of considering the applicability domain (AD) of the model. Several AD analyses including the descriptor ’s range, the applicability domain (AD) of the model. Several AD analyses including the descriptor’s range, leverages, Euclidian distances, rivality and modelability indexes [66–68] are available. In this study we leverages, Euclidian distances, rivality and modelability indexes [66–68] are available. In this study used two of AD approaches; maximal Euclidian distance of training set in NN-C (EDcrt = 6.16) and we used two of AD approaches; maximal Euclidian distance of training set in NN-C (EDcrt = 6.16) NN-D model (ED = 2.36), and leverage values in Q-D model (hat* = 0.4) (Figure 4) [67,68]. crt and NN-D model (EDcrt = 2.36), and leverage values in Q-D model (hat* = 0.4) (Figure 4) [67,68]. Molecules 2019, 24, 837 7 of 21 Molecules 2019, 24, x FOR PEER REVIEW 6 of 21 Figure 4. Graphical assessment of the applicability domains of the best three classiﬁcation models Figure 4. Graphical assessment of the applicability domains of the best three classification models (NN-C, NN-D, Q-D). Compounds with estimated Euclidian distances/leverages > ED /hat* fail out (NN-C, NN-D, Q-D). Compounds with estimated Euclidian distances/leverages > EDcrit/hat* fail out crit of applicability domain of model. of applicability domain of model. The structure activity relationship analysis has revealed the inﬂuential structural properties of BTL The structure activity relationship analysis has revealed the influential structural properties of ligands. In general, high capability of compounds of forming hydrogen bonds, ionic properties and BTL ligands. In general, high capability of compounds of forming hydrogen bonds, ionic properties planarity of molecules are the most relevant one. The same features were highlighted in previous BTL and planarity of molecules are the most relevant one. The same features were highlighted in experimental studies [53,54,59]. The molecular descriptors, which are used in newly developed previous BTL experimental studies [53,54,59]. The molecular descriptors, which are used in newly models NN-D and Q-D (Table S4, Supplementary Material) represent the following properties: developed models NN-D and Q-D (Table S4, Supplementary Material) represent the following planarity (number of multiple and aromatic bonds — nBM, nArCO, nArOR, ARR), hydrogen bonds properties: planarity (number of multiple and aromatic bonds — nBM, nArCO, nArOR, ARR), forming ability (nOHp, nO, H-052, P_VSA_LogP_2, P_VSA_p_4, CATS2D_07_DA, CATS2D_04_AN), hydrogen bonds forming ability (nOHp, nO, H-052, P_VSA_LogP_2, P_VSA_p_4, CATS2D_07_DA, ionic properties (MATS7e, GATS5i, EE_B(s)). CATS2D_04_AN), ionic properties (MATS7e, GATS5i, EE_B(s)). 2.1.1. Consensus of Model Predictions Table 1. Statistic parameters of the best three single models and consensus classification models. Consensus modeling was performed by integrated predictions among two QSAR models (2-model Single Models 2-Model Consensus 3-Model Consensus consensus, three possible combinations) and all three models together (3-model consensus). In general, NN-C NN-D Q-D NN-C + NN-D NN-C + Q-D NN-C + Q-D A B A + B researchers acknowledged an increase in overall predictive performance, if consensus of predictions TP 44 44 46 33 33 41 29 19 48 among a battery of models was applied [69]. Following the protocol of Marzo et al. [69] we made TN 61 65 58 47 49 55 45 17 62 various combinations of consensus of model predictions by combining predictions from NN-C, FP 9 5 12 6 2 1 1 7 8 NN-D and Q-D models. The 3-model consensus A was the strictest and considered only predictions FN 6 6 4 0 2 0 0 2 2 with∑ all thr120 120 120 ee models (NN-C + NN-D +86 Q-D) in agreement. 86 In 3-model 97 consensus 75 45 ‘A + B’120 the PR 1 (0.98) 1 (0.92) 1 (0.97) 0.72 0.72 0.81 0.62 0.37 1 predictions with maximal one disagreement among the NN-C, NN-D and Q-D models were considered. SE 0.88 0.88 0.92 1.00 0.94 1.00 1.00 0.90 0.96 In 2-model consensus ‘NN-D + Q-D’ the predictions where both models were in agreement were SP 0.87 0.93 0.83 0.89 0.96 0.98 0.98 0.71 0.89 considered. In the same way the rest two combinations of a 2-model consensus (NN-C + NN-D and ACC 0.88 0.91 0.87 0.93 0.95 0.99 0.99 0.80 0.92 NN-C + Q-D) were also performed, but have lowest performance evaluation than ‘NN-D + Q-D’. NPV 0.91 0.92 0.94 1.00 0.96 1.00 1.00 0.89 0.97 The results of performance evaluation of consensus modeling are represented in Tables 1 and 2, and PPV 0.83 0.90 0.79 0.85 0.94 0.98 0.97 0.73 0.86 the detailed predictions in Table S1 (Supplementary Material). When compared with the single models, MCC 0.75 0.81 0.74 0.87 0.90 0.98 0.97 0.62 0.84 the accuracy performance of consensus of predictions was increased, but consequently the prediction A—complete agreement among predictions, B—maximal one disagreement among predictions. rate was decreased. The highest overall accuracy (0.99) was determined in 2-model consensus ‘NN-D + Q-D’ and 3-model consensus A, yet both consensus signiﬁcantly differ in prediction rate ability, 81% and 62%, respectively. Therefore, the best balance among accuracy and prediction rate was achieved with 3-model consensus ‘A + B’ (ACC = 0.92, PR = 100%). This 3-model consensus was shown as robust and reliable, and in evaluation parameters outperformed all single models. In this regard, the high impact of integration of predictions of three models was conformed and in concordance with expectations the overall predictive performance was increased. 2.1.2. Application of Classiﬁcation Models on Phenolic Dataset Developed QSAR models enabled efﬁcient application in screening processes regarding bioavailability of dietary/pharmaceutically interesting compounds. With single classiﬁcation models and 2-/3-model consensus of predictions we estimated the BTL transport potential for compounds, which were not experimentally tested for BTL transport activity (phenolic dataset, n = 300). Results of Molecules 2019, 24, x FOR PEER REVIEW 8 of 21 Among single models the highest number of active compounds was predicted with the model NN-D (138), which was significantly higher than in other models (NN-C = 75, Q-D = 72). However, none of single models or consensus of predictions did not recognize groups of phenols that are more likely to interact with BTL (Figure 5). For sure the most promising active compounds are those 15 that were predicted in all models: luteolin (ID4), kaempferol (ID86), eriodictyol (ID95), pinobanksin (ID117), cianidanol (ID127), leucodelphinidin (ID131), ellagic acid (ID181), rosmarinic acid (ID182), gallic acid (ID199), methyl gallate (ID200), 3-methoxy-4-hydroxybenzoic acid (ID209), 3-methoxy-4-hydroxyhippuric acid (ID211), decanyl caffeate (ID225), oleuropein (ID226), PACD3 (ID280) (see Table S2, column A, Supplementary Material). Since in vivo/in vitro experimental data on membrane transport ability for majority of the above mentioned compounds are not available either for BTL neither for other transporters from the list in Table 3 (except for four compounds and Molecules 2019, 24, 837 8 of 21 few transporters, see results in subsection 2.3.), the in silico predictions are currently the only relevant estimations of their membrane transport activities. Interestingly, based on our results BTL might be also involved in the gastric absorption of the flavone luteolin, as its gastric uptake was predictions are represented in Figure 5 and Table S2 (Supplementary Material). Consensus ‘A + B’ observed in vivo [70]. Indeed, this is in support of hypothesis of Passamonti et al. [35] that at the and single models N-C and Q-D performed with a 100% prediction rate with most of the compounds gastric level the BTL-mediated mechanism could play an important role in active transport of within AD (A + B = 300, NN-C = 283, Q-D = 278). On the other hand, the model NN-D resulted polyphenols. The isoflavone daizdein, on the other hand, was classified as inactive with our with a lower number of compounds in AD (NN-D = 208). As expected, lower prediction rate was consensus of QSAR models and was also experimentally identified as not being a BTL substrate [35]. evaluated for other consensus of predictions (NN-D + Q-D = 50%, A = 36%), due to strictest conditions. However, study of Piskula et al. [71] demonstrated that daizdein can be absorbed from the rat Generally, the integration of multiple models increased the overall reliability of predictions in all stomach, thus in its gastric uptake probably another specific OAT transporter, a monocarboxylate consensus combinations, also increased the prediction rate for phenolic compounds in consensus transporter, is involved [35]. ‘A + B’, but decreased in other consensus (NN-D + Q-D, A). Figure 5. Representation of classiﬁcation of 300 compounds with three separate classiﬁcation models Figure 5. Representation of classification of 300 compounds with three separate classification models (NN-C, NN-D, Q-D) and three consensus models (A + B, NN-D + Q-D, A) on graphical map. (NN-C, NN-D, Q-D) and three consensus models (A + B, NN-D + Q-D, A) on graphical map. Using in silico models one is challenged with the paradigm of selecting single model or very strict consensus (e.g., A) with high accuracy and narrow AD, or on the price of broadening of AD decide for wider consensus (e.g., A + B). In this regard, the number of active compounds predictions varied from 15 in consensus A to 65 in consensus ‘A + B’ (Table S2, Supplementary Material). Among single models the highest number of active compounds was predicted with the model NN-D (138), which was signiﬁcantly higher than in other models (NN-C = 75, Q-D = 72). However, none of single models or consensus of predictions did not recognize groups of phenols that are more likely to interact with BTL (Figure 5). For sure the most promising active compounds are those 15 that were predicted in all models: luteolin (ID4), kaempferol (ID86), eriodictyol (ID95), pinobanksin (ID117), cianidanol (ID127), leucodelphinidin (ID131), ellagic acid (ID181), rosmarinic acid (ID182), gallic acid (ID199), methyl gallate (ID200), 3-methoxy-4-hydroxybenzoic acid (ID209), 3-methoxy-4-hydroxyhippuric acid (ID211), decanyl caffeate (ID225), oleuropein (ID226), PACD3 (ID280) (see Table S2, column A, Supplementary Material). Since in vivo/in vitro experimental data on membrane transport ability for majority of the above mentioned compounds are not available either for BTL neither for other transporters from the list in Table 3 (except for four compounds and few transporters, see results in Section 2.3), the in silico predictions are currently the only relevant estimations of their membrane transport activities. Interestingly, based on our results BTL might be also involved in the gastric absorption of the ﬂavone luteolin, as its gastric uptake was observed in vivo [70]. Indeed, this is in support of hypothesis of Passamonti et al. [35] that at the gastric level the BTL-mediated mechanism could play an important role in active transport of polyphenols. The isoﬂavone daizdein, on the other hand, was classiﬁed as inactive with our consensus of QSAR models and was also experimentally identiﬁed as not being a BTL substrate [35]. However, study of Piskula et al. [71] demonstrated that daizdein can be absorbed from the rat stomach, thus in its gastric uptake probably another speciﬁc OAT transporter, a monocarboxylate transporter, is involved [35]. Molecules 2019, 24, 837 9 of 21 Table 3. List of transporters from Metrabase and ChemBench platforms. Protein Symbol Gene Symbol Name Tissue Expression Data M,C MDR1 ABCB1 ATP-binding cassette sub-family B member 1, P-glycoprotein intestine, kidney, liver, brain BSEP ABCB11 ATP binding cassette subfamily B member 11 liver M,C MRP1 ABCC1 ATP-binding cassette, sub-family C member 1 intestine, kidney, liver, brain M,C ABCC2 ATP-binding cassette sub-family C member 2 stomach, intestine, kidney, liver, brain MRP2 M,C MRP3 ABCC3 ATP-binding cassette sub-family C member 3 stomach, intestine, kidney, liver, brain M,C ABCC4 ATP-binding cassette, sub-family C member 4 stomach, intestine, kidney, liver, brain MRP4 MRP5 ABCC5 ATP Binding Cassette Subfamily C member 5 kidney, brain M,C BCRP1 ABCG2 ATP-binding cassette sub-family G member 2 stomach, intestine, kidney, liver, brain SLCO1A2 solute carrier organic anion transporter family member 1A2 kidney, liver, brain OATP1A2 OATP2A1 SLCO2A1 solute carrier organic anion transporter family member 2A1 intestine, kidney, liver, brain OATP1B1 SLCO1B1 solute carrier organic anion transporter family member 1B1 liver OATP1B3 SLCO1B3 solute carrier organic anion transporter family member 1B3 liver M,C SLCO2B1 solute carrier organic anion transporter family member 2B1 intestine, kidney, liver, brain OATP2B1 OATP3A1 SLCO3A1 solute carrier organic anion transporter family member 3A1 intestine, kidney, liver, brain OATP4A1 SLCO4A1 solute carrier organic anion transporter family member 4A1 stomach, intestine, kidney, liver, brain GLUT1 SLC2A1 solute carrier family 2 (facilitated glucose transporter) member 1 intestine, kidney, liver, brain NTCP SLC10A1 solute carrier family 10 member 1 liver ASBT SLC10A2 solute carrier family 10 Member 2 intestine SLC15A1 solute carrier family 15 member 1 intestine PEPT1 M,C MCT1 SLC16A1 solute carrier family 16 member 1, monocarboxylic acid transporter 1 stomach, intestine, kidney, liver, brain M,C OCT1 SLC22A1 solute carrier family 22 (organic cation transporter) member 1 intestine, kidney, liver BTL * / bilitranslocase, bilirubin membrane transporter stomach, intestine, kidney, liver * new, not available in Metrabase and ChemBench, M—available in Metrabase, C—available in ChemBench. Molecules 2019, 24, 837 10 of 21 2.2. BTL Regression Models The activity predictions for the BTL inhibition constants pKi [mmol/L] were performed with the best in-house predictive CP-ANN models (M1–M4) [59]. The neural networks were selected regarding results from the previous studies [59]. Firstly, we analyzed, if the regression models were accurately developed and validated regarding the need of evenly frequency distribution of compounds in all ranges of BTL biological activity in the training and validation datasets. This condition was fully satisﬁed and a compounds frequency distribution in ﬁve intervals of BTL activity is shown in Figure S1 (Supplementary Material). For further detailed evaluation of models a ﬁve quality parameters of 2 2 2 validation were employed (R , RMSE, Q , CCC, r ), and their performance is represented in Table 4. F3 m Statistical differences among models are low and in general all validation parameters characterize model M1 as the best one. The validation parameter r was specially reported to help in identifying the best models among a set of comparable models [72], but in this case all parameters were equally sensitive. Since GA approach was used in the development and optimization of models, consequently a large availability of different models predicting the same endpoint was produced. Thus, to avoid having cluster of similar models we tested the diversity of the top scoring models by measuring of Hamming distance between them. The results are shown in Table S5 (Supplementary Material). The distances among models are in range of 7–17. This indicated that numbers of variables not shared by the models are high. Accordingly, our set of top models was shown to be diverse and consensus analyses were justiﬁed. An average response and weighted average response were used as proposed by Todeschini et al. [65]. The consensus gave better statistical ﬁts of predictions in regard to the single models. Table 4. Statistic parameters of the best four regression models. 2 2 2 2 2 2 Model R RMSE R Q CCC R Q CCC RMSE CCC r tr ev ev m iv iv F3iv iv F3ev all all (all) M1 0.996 0.266 0.949 0.957 0.970 0.872 0.927 0.939 0.198 0.987 0.821 M2 0.999 0.280 0.943 0.953 0.968 0.842 0.909 0.927 0.206 0.986 0.812 M3 0.998 0.289 0.939 0.950 0.964 0.839 0.908 0.934 0.213 0.985 0.806 M4 0.995 0.295 0.937 0.948 0.964 0.760 0.863 0.899 0.243 0.980 0.776 Application of Regression Models on Phenolic Dataset Predictions were made for 205 compounds. These were compounds that were classiﬁed as active with at least one model (Table S7, Supplementary Material). In order to produce more reliable estimates of our endpoint the consensus modeling strategy was also applied. Most of models were well applicable for phenols, since majority of compounds were in AD of models (>75%); exception was model M3 with only 50% of compounds in AD. Moreover, the predictions were highly harmonized, since varied for less than 10% of standard deviation, if compared among all models (M1-M4). Up to our knowledge only one study exist on extrapolation of BTL models to make predictions of new compounds, which were not included in model development or validation [59]. Comparison with that study revealed that BTL models have wider capacity for prediction of phenolic compounds than other antioxidants and antiprions (Table S6, Supplementary Material). 2.3. Metrabase and Chembench Analyses Based on data and tools available on Metrabase and Chembench platforms we made the comprehensive virtual proﬁling of membrane transport interactions for compounds of both datasets. Herein, we performed the unique transporter proﬁle of studied compounds, which is based on the results from new BTL models and hits/predictions from over 20 membrane transporters of Metrabase and Chembench (Figures 6 and 7). Molecules 2019, 24, x FOR PEER REVIEW 11 of 21 37–84), nucleobases, nucleosides and nucleotides (ID 1–36). The only exception was few flavonoids (ID 65–70). On the other hand, drugs and endogenous substances (ID 85–120), which are of interest for pharmaceutical industry, were better studied and experimentally analyzed, therefore more hits Molecules 2019, 24, 837 11 of 21 were found. non-inhibitor inhibito ambigious expression MDR1 MRP1 MRP2 MRP3 MRP4 BCRP1 GLUT1 MCT1 OCT1 OATP1A2 OATP1B1 OATP1B3 OATP2A1 OATP2B1 OATP3A1 OATP4A1 BTL nucleobase nucleoside nucleotide phenols drug other Figure 6. Activity proﬁle for 120 compounds including BTL experimental data [53,54] and data Figure 6. Activity profile for 120 compounds including BTL experimental data [53,54] and data Molecules 2019, 24, x FOR PEER REVIEW 12 of 21 available for 16 transporters in Metrabase [46]. Grey—no data available in database. available for 16 transporters in Metrabase [46]. Grey—no data available in database. Based on these results we can conclude that for phenolic compounds, nucleobases, nucleosides and nucleotides the only available information regarding experimental transport activity of the membrane transporters is those of BTL provided from research group of prof. Passamonti [53,54], which are not included in Metrabase. Further on, we made analyses in Metrabase for compounds of consensus A, which were predictive as active (15 compounds, see Subsection 2.1.2). We found out that only for luteolin (ID4), kampferol (ID86), cianidanol (ID127) and eriodictyol (ID95) a few experimental studies exist in this database; luteolin (MDR1–inhibitor, BCRP1–inhibitor, MRP2– non-inhibitor), kaempferol (BCRP1–inhibitor, MRP2–non-inhibitor, OCT1–inhibitor, OATP1A2– inhibitor), cianidanol (MRP2–non-inhibitor or substrate, OCT1–non-inhibitor, OATP2B1–inhibitor or non-inhibitor) eriodictyol (MRP2–non-inhibitor). Besides this four compounds, which were captured the significant attention of the scientific audience for their potential anti-oxidative, anti-tumor, and anti-inflammatory effects, also other phenolic compound of natural origin like oleuropein (ID226) and leucodelphinidin (ID131) are of interest due to their anti-neurodegenerative and hypolipidaemic effects [73,74]. Thus, the predictions about transport ability by BTL and 14 others importers and exporters that are represented in Tables S2 and S9 (Supplementary Material) are of high interest for clarification about their active transport across membrane. Further on, we made predictions for 420 compounds (BTL and phenolic datasets) by using QSAR models of 14 transporters available in Human intestinal transporter database at Chembench platform [44]. The results of predictions are represented in Figure 7, Table S8 and S9 (Supplementary Material). In general, the models have been found to have good prediction ability and coverage for phenolic compounds, since only few compounds were out of AD or unable to predict (Table S9, Supplementary Material). Yet, almost all models had poor prediction ability for nucleobases, nucleosides and nucleotides (ID 1–36, Table S8, Supplementary Material). Poor prediction ability was observed also for some flavonoids in models of MRP4, MRP5, MCT1, NTCP, ASBT and OCT1 (anthocyanidins (ID 71–84) and flavonols (ID 52–64)). Figure 7. Activity proﬁle of computational predictions for 14 transporters from ChemBench and for Figure 7. Activity profile of computational predictions for 14 transporters from ChemBench and for our BTL model, red—active, blue—inactive, grey—no prediction; (a) BTL dataset of 120 compounds, our BTL model, red—active, blue—inactive, grey—no prediction; (a) BTL dataset of 120 compounds, (b) phenols dataset of 300 compounds. (b) phenols dataset of 300 compounds. We compared the transport activity data among BTL and 16 transporters that are publicly 3. Discussion available in Metrabase (cheminformatics and bioinformatics database) [46]. Analysis of hits of BTL In spite a lack of adequate experimental validation for our in silico predictions, some dataset (compounds from Table S1, Supplementary material) in Metrabase database is represented information could be gained by comparing structurally similar compounds. Among the fifteen the on Figure 6. It was found out that almost no experimental data exist for the phenolic compounds most promising active compounds, which were selected with the BTL QSAR models, luteolin, kaempferol, eriodictyol and pinobanksin were found to share the highest structural similarity with quercetin. Quercetin, malvidin 3-glucoside, bilirubin and bromosulfophthalein are up to date the only compounds identified experimentally as BTL substrates [55,75]. Hopefully, the results published in this study would be inspiration for in vitro experts to test the most promising predicted active compounds with the BTL transport activity assay. The compounds of natural origin with high health benefit should be selected first, like oleuropein, luteolin, eriodictyol, cianidanol, and leucodelphinidin. The clinic reports highlighted the impact of knowledge about how various therapeutics or dietary supplements interacts with membrane transporters [76]. Herein, for the first time were represented joint analyses of transport activities of BTL and several others membrane transporters. Results not solely present a new insight in interactions between phenols and several membrane transporters, but can help also in assessment of their potential absorption and disposition properties, yet gain also valuable information for understanding possible interference of phenols with the transport activity of drugs. Since ingestion of extra dietary phenolic supplements as support to chemotherapy was proposed as promising therapeutic approach [7,10], but such diet could consequently influence drugs absorption and disposition [58]. For proper drug application this is for Molecules 2019, 24, 837 12 of 21 (ID 37–84), nucleobases, nucleosides and nucleotides (ID 1–36). The only exception was few ﬂavonoids (ID 65–70). On the other hand, drugs and endogenous substances (ID 85–120), which are of interest for pharmaceutical industry, were better studied and experimentally analyzed, therefore more hits were found. Based on these results we can conclude that for phenolic compounds, nucleobases, nucleosides and nucleotides the only available information regarding experimental transport activity of the membrane transporters is those of BTL provided from research group of prof. Passamonti [53,54], which are not included in Metrabase. Further on, we made analyses in Metrabase for compounds of consensus A, which were predictive as active (15 compounds, see Section 2.1.2). We found out that only for luteolin (ID4), kampferol (ID86), cianidanol (ID127) and eriodictyol (ID95) a few experimental studies exist in this database; luteolin (MDR1–inhibitor, BCRP1–inhibitor, MRP2–non-inhibitor), kaempferol (BCRP1–inhibitor, MRP2–non-inhibitor, OCT1–inhibitor, OATP1A2–inhibitor), cianidanol (MRP2–non-inhibitor or substrate, OCT1–non-inhibitor, OATP2B1–inhibitor or non-inhibitor) eriodictyol (MRP2–non-inhibitor). Besides this four compounds, which were captured the signiﬁcant attention of the scientiﬁc audience for their potential anti-oxidative, anti-tumor, and anti-inﬂammatory effects, also other phenolic compound of natural origin like oleuropein (ID226) and leucodelphinidin (ID131) are of interest due to their anti-neurodegenerative and hypolipidaemic effects [73,74]. Thus, the predictions about transport ability by BTL and 14 others importers and exporters that are represented in Tables S2 and S9 (Supplementary Material) are of high interest for clariﬁcation about their active transport across membrane. Further on, we made predictions for 420 compounds (BTL and phenolic datasets) by using QSAR models of 14 transporters available in Human intestinal transporter database at Chembench platform [44]. The results of predictions are represented in Figure 7, Tables S8 and S9 (Supplementary Material). In general, the models have been found to have good prediction ability and coverage for phenolic compounds, since only few compounds were out of AD or unable to predict (Table S9, Supplementary Material). Yet, almost all models had poor prediction ability for nucleobases, nucleosides and nucleotides (ID 1–36, Table S8, Supplementary Material). Poor prediction ability was observed also for some ﬂavonoids in models of MRP4, MRP5, MCT1, NTCP, ASBT and OCT1 (anthocyanidins (ID 71–84) and ﬂavonols (ID 52–64)). 3. Discussion In spite a lack of adequate experimental validation for our in silico predictions, some information could be gained by comparing structurally similar compounds. Among the ﬁfteen the most promising active compounds, which were selected with the BTL QSAR models, luteolin, kaempferol, eriodictyol and pinobanksin were found to share the highest structural similarity with quercetin. Quercetin, malvidin 3-glucoside, bilirubin and bromosulfophthalein are up to date the only compounds identiﬁed experimentally as BTL substrates [55,75]. Hopefully, the results published in this study would be inspiration for in vitro experts to test the most promising predicted active compounds with the BTL transport activity assay. The compounds of natural origin with high health beneﬁt should be selected ﬁrst, like oleuropein, luteolin, eriodictyol, cianidanol, and leucodelphinidin. The clinic reports highlighted the impact of knowledge about how various therapeutics or dietary supplements interacts with membrane transporters [76]. Herein, for the ﬁrst time were represented joint analyses of transport activities of BTL and several others membrane transporters. Results not solely present a new insight in interactions between phenols and several membrane transporters, but can help also in assessment of their potential absorption and disposition properties, yet gain also valuable information for understanding possible interference of phenols with the transport activity of drugs. Since ingestion of extra dietary phenolic supplements as support to chemotherapy was proposed as promising therapeutic approach [7,10], but such diet could consequently inﬂuence drugs absorption and disposition [58]. For proper drug application this is for sure of high importance. Moreover, results of the systematic analysis performed in this study are also important for elucidation of functional role of BTL, as up to our knowledge only limited data exist. Up to date, only one Molecules 2019, 24, 837 13 of 21 synopsis of reported interactions with BTL and other membrane transporters one a few compounds (>50) is available [58]. Since in general not much is known about interactions between phenols and transporters, the results from this study are of interest [77]. Moreover, we managed to conﬁrm our expectations, that newly developed BTL QSAR models have good predictive ability and coverage for phenolic compounds. In future it would be of interest to make a similar study on the pharmaceutically interested nucleobase derivatives [78–80], for which it is assumed that BTL could have important role on their transport [54]. More than half a century has passed since an implementation of Hansch-Fujita approach in the formulation of QSAR, which provided a new perspective for chemical-biological interactions and a number of successful applications are known especially in drug discovery [77]. Nowadays, we have abundance of QSAR models for predictions of numerous endpoints in various ﬁelds of science. In this study we tried to collect and compare the QSAR models of membrane transporters to make a comprehensive transporter proﬁles for purposes of virtual screening of compounds for their bioavailabilty. Now point of future investigation should be to switch in new level beyond individual approaches and integrate various QSAR analyses with ontology and omics data [77,81–83]. Such simultaneously predicting with joint data analyses are shown to increase the accuracy of models by exploiting their common representation and identifying common features between individual properties, which are frequently strongly correlated with one another [81,83]. For sure not in so far future a chemical characterization will be even more empowered with an artiﬁcial intelligence assisted software, which would make decisions based on integration of numerous data from in silico, in vitro, in vivo and in situ analyses [84]. Thus, hopefully models from this study will serve like small piece of a large puzzle in the complex biochemical system analysis. 4. Materials and Methods Quantitative structure-activity relationships (QSARs) are widely used in silico methods to predict biological activity for untested chemicals (Figure 1). This is proven, time and cost-effective approach, which is in most cases sufﬁcient alternative method to animal testing [44]. For building and validation of models the appropriate statistical algorithms and the data matrix which includes numerical values of chemical structures and empirical values of biological activity are needed. 4.1. Dataset The dataset for modeling included 120 aromatic organic compounds (43 phenols and 77 various aromatic compounds like nucleobase derivatives, drugs, dyes, etc.) [59]. In general chemical structures of all compounds were optimized and numerically coded with molecular descriptors (MD). The MD calculation and selection of inﬂuential MD is described in the Section 4.2.1. The corresponding empirical data of BTL transport activity assay were collected from experiments of Župerl et al. [54] and Karawajczk et al. [53]. Inhibition constants (K ) were determined based on Michaelis-Menten constants, which were obtained from spectrophotometrical evaluation of BTL transport activity on selected compounds in medium of sulfobromophtalein (BSP) and rat liver plasma membrane vesicles. According to threshold criteria proposed by Župerl et al. [54] the dataset composed of 50 active compounds (inhibitors, pK > 1.3) and 70 inactive compounds (non-inhibitors, pK 1.3). The compounds are listed in Table S1 (Supplementary Material). The test dataset was compiled from literature. It encompasses natural occurring and synthetic phenolic compounds having various potential health beneﬁts. Altogether 300 compounds were collected from over 30 publications [7,11–30,85–93], see Table S2 (Supplementary Material). Compounds were arranged according to their structural similarity into 12 groups: ﬂavones/isoﬂavones, ﬂavonols, ﬂavanones/ isoﬂavanones, ﬂavanonols, ﬂavanols, xanthones, chromones, coumarins, chalcones, phenolic acid derivatives, simple phenols and other compounds. Experimental evaluation of potential role of selected phenolic compounds in preventing various chronic diseases are explained in detail in the references listed in Table S2 (Supplementary Material). Molecules 2019, 24, 837 14 of 21 4.2. Computational Analysis 4.2.1. Molecular Descriptors Molecular structures were represented with a set of molecular descriptors (MDs). First, the MOPAC software [94] was used for optimization of the 2D chemical structures into 3D conformations considering the minimal energy criterion for the optimal conformation of the compound in vacuum. Then numerical coding of 3D conformations was performed with MDs, which were calculated with the CODESSA program [95]. With the applied software more than 400 MDs for each compound were calculated. Second, the numerical encoding of 2D conformations was performed with Dragon 7.0 software (Kode srl., Pisa, Italy) [96], which calculated over 3000 MDs for each molecule. The dataset with Codessa and Dragon descriptors were further on treated separately. So, obtained two datasets were normalized to zero mean and unit standard deviation for each descriptor. Afterwards, the reduction of numerous descriptors was performed by omitting the MDs with zero variance or high intercorrelation (>0.9). The set of descriptors was then reduced according to the similarity criterion in the Kohonen map by following the protocol of Martinci ˇ c ˇ et al. [59], which resulted in the ﬁnal data pool of 66 and 87 MDs for Codessa and Dragon dataset, respectively. 4.2.2. Generation of Models The inputs for model building were various m-dimensional vectors representing the chemical structure, m being the number of MDs selected (independent variables), and the targets (properties, i.e. dependent variables) corresponding to pK of BTL transport activity. We used CPANNatNIC software for the splitting of data for modeling into training and validation set [59,92]. For model ﬁtting the linear and nonlinear regression approaches were used. The multiple linear regression models were developed with Qsarins software (DiSTA, Varese, Italy, www.qsar.it) [68]. The supervised learning models were developed by using counter-propagation artiﬁcial neural network (CP-ANN) algorithm and in-house software [97,98]. For classiﬁcation and prediction of unknown compounds two QSAR modeling strategies were adopted. Firstly, the classiﬁcation models were developed to separate chemicals into active/non-active classes, and then optimized regression models were used for the active chemicals to predict the values of the inhibition constants [59]. All 120 compounds were considered for the construction and optimization of the classiﬁcation model, while the predictive models are based on active compounds only. For optimization we applied cross-validation (CV) and genetic algorithm (GA) [65,98] in order to select the inﬂuential descriptors and to improve the predictive ability and robustness of the models. In models selection the applicability domain (AD) [67] was applied to evaluate the model predictions within the established chemical space limits. 4.2.3. Statistical Evaluation of Models The criteria for selection of the best classiﬁcation models, so called quality indicators, are listed in Table 5 [99,100]. Models with the highest quality indicators of the validation set and wider applicability (larger AD) were further analyzed. Two approaches were used for AD evaluation, the cumulative distributions of Euclidian distances to central neurons (MEDS) for CP-ANN models [67] and the leverage values for MLR models [68]. Several models were built by each of the selected modelling method. Further on, the consensus of model predictions was performed by combining predictions of three or two QSAR models. The consensus was performed following the protocol of Marzo et al. [69], which including arranging of compounds in two types: A—complete agreement among predictions, B-maximum one disagreement among the predictions. Molecules 2019, 24, 837 15 of 21 Table 5. Quality indicators for evaluation of different classiﬁcation models. Acronym Quality Indicator Formula TP true positive TN true negative FP false positive FN false negative PR predict ratio predicted compounds/total compounds SE, TPR sensitivity, true positive rate TP/(TP + FN) SP, TNR speciﬁcity, true negative rate TN/(TN + FP) ACC accuracy (SP + SE)/2 NPV negative predictive value TN/(TN + FN) PPV positive predictive value, precision TP/(TP + FP) ((TP*TN)-(FP*FN))/ [(TP + FP)(TP + MCC Matthews correlation coefﬁcient FN)(TN + FP)(TN + FN)] The criteria for selection of the best regression models were several validation parameters 2 2 2 (R , RMSE, CCC, Q , r ) [64,70]. The consensus approach was also applied for the best four F3 m regression models (M1-M4) by using the weighted average response. The calculation was performed following the Equation (1); M—number of models, y —response estimated by the k-th model, h —leverage [65]: M k k=1 h y = (1) M 1 k=1 In order to evaluate diversity of models the Hamming distances d between models were accounted based on the two-way table [65] (Equation (2)). The table represent the relationship between two binary vectors, each vector represent one model; b—the number of cases that for same position there is 1 in vector I and 0 vector II, c—the number of cases that for same position there is 0 in vector I and 1 vector II Values in Table S5 (Supplementary Material) represent the number of MD not shared among models: d = b + c (2) 4.3. Metrabase and Chembench Analyses The public available cheminformatics and bioinformatics database for small molecule transporter data analysis (Metrabase) [46,101] was screened for transport activity proﬁle of 120 compounds (BTL dataset). The synopsis of transport activity among BTL and 16 transporters from Metrabase was performed. The selected 16 transporters are listed in Table 3. Further on, in Human intestinal transporter database, available on Chembench platform, the QSAR models of 14 transporters were used to make predictions for 420 compounds (BTL and phenolic datasets) [44,102]. The transporters are listed in Table 3 and detailed list of QSAR models properties is summarized in Table S3 (Supplementary Material). The comparison of transport activity among BTL and 14 transporters from Chembench was performed for both datasets. 5. Conclusions New classiﬁers for BTL transport activity were developed. Based on in silico approach for estimation of BTL transport activity the classiﬁcation for dietary/pharmaceutically interesting compounds (natural dietary phenols and synthetic phenol-like drugs) were performed. Using QSAR classiﬁcation models (NN-C, NN-D, Q-D) we estimated the BTL transport activity of 300 compounds, which have not been previously experimentally evaluated. Moreover, the consensus among battery of models was applied, since the detailed evaluation of models shown better performance than individual models. With various predictive CP-ANN models (M1-M4) then the transport activity prediction (inhibition constant pKi [mmol/L]) for the compounds classiﬁed as active was assessed. The reliability Molecules 2019, 24, 837 16 of 21 of all predictions was evaluated with the applicability domain (AD) of each individual classiﬁcation and regression model. In general, BTL is known as a potential delivery system for wide variety of the vitally important endogenous substrates and drugs, yet with the models represented in this study a good ability for predicting phenolic compounds was shown too. The 205 compounds from the dataset of 300 phenols have a potential to interact with BTL. The central part of this study was the synopsis of unique transporter proﬁles of BTL in comparison with 20 other membrane transporters from Metrabase and Chembench. Results present a new insight in interactions between phenols and several membrane transporters, which can help not solely in assessment of their potential absorption and disposition properties but also gain insight in possible interference of phenols with the transport activity of drugs, what can be valuable information in drug application. Supplementary Materials: The list of ChemBench models and detailed list of BTL and phenolic datasets with extended results of data analyses (Tables S1–S9) are available as supplementary material online. Author Contributions: Conceptualization and design of study, K.V. and M.N.; data curation, K.V.; calculations, data analysis and validation, K.V.; writing—original draft preparation and editing, K.V.; writing—review, discussion, M.N.; project administration, M.N.; funding acquisition, M.N. Funding: This work was ﬁnancially supported by the grants from the Slovenian Research Agency (national program P1-0017 and research project J3-8209). Acknowledgments: We thank Viktor Drgan for technical support and Paola Gramatica for availability of Qsarins license. Conﬂicts of Interest: The authors declare no conﬂict of interest. References 1. Del Rio, D.; Rodriguez-Mateos, A.; Spencer, J.P.; Tognolini, M.; Borges, G.; Crozier, A. Dietary (poly)phenolics in human health: Structures, bioavailability and evidence of protective effects against chronic diseases. Antioxid. Redox. Signal. 2013, 18, 1818–1892. [CrossRef] [PubMed] 2. Dai, J.; Mumper, R.J. Plant Phenolics: Extraction, Analysis and Their Antioxidant and Anticancer Properties. Molecules 2010, 15, 7313–7352. [CrossRef] [PubMed] 3. Harborne, J.B. Plant phenolics. In Encyclopedia of Plant Physiology, Volume 8: Secondary Plant Products; Bell, E.A., Charlwood, B.V., Eds.; Springer: Berlin, Germany; New York, NY, USA, 1980; pp. 329–395. 4. Khoddami, A.; Wilkes, M.A.; Roberts, T.H. Techniques for analysis of plant phenolic compounds. Molecules 2013, 18, 2328–2375. [CrossRef] [PubMed] 5. Nicholson, R.L.; Vermerris, W. Phenolic compound biochemistry; Springer: Berlin, Germany, 2006; pp. 107–108. 6. FooDB version 1.0. Available online: www.foodb.ca (accessed on 15 October 2018). 7. Fresco, P.; Borges, F.; Diniz, C.; Marques, M.P. New Insights on the Anticancer Properties of Dietary Polyphenols. Med. Res. Reviews 2006, 26, 747–766. [CrossRef] [PubMed] 8. Pietta, P.G. Flavonoids as antioxidants. J. Nat. Prod. 2000, 63, 1035–1042. [CrossRef] [PubMed] 9. Lu, Z.; Nie, G.; Belton, P.S.; Tang, H.; Zhao, B. Structure-activity relationship analysis of antioxidant ability and neuroprotective effect of gallic acid derivatives. Neurochem. Internat. 2006, 48, 263–274. [CrossRef] [PubMed] 10. Rasines-Perea, Z.; Teissedre, P.L. Grape Polyphenols’ Effects in Human Cardiovascular Diseases and Diabetes. Molecules 2017, 22, 68. [CrossRef] [PubMed] 11. Khalil, M.I.; Sulaiman, S.A. The potential role of honey and its polyphenols in preventing heart diseases: A review. Afr. J. Tradit. Complement. Altern. Med. 2010, 7, 315–321. [CrossRef] [PubMed] 12. Machado, N.F.L.; Calheiros, R.; Fiuza, S.M.; Borges, F.; Gaspar, A.; Garrido, J.; Marques, M.P. Phenolic esters with potential anticancer activity—The structural variable. J. Mol. Model. 2007, 13, 865–877. [CrossRef] [PubMed] 13. Bauvois, B.; Puiffe, M.L.; Bongui, J.B.; Paillat, S.; Monneret, C.; Dauzonne, D. Synthesis and Biological Evaluation of Novel Flavone-8-acetic Acid Derivatives as Reversible Inhibitors of Aminopeptidase N/CD13. J. Med. Chem. 2003, 46, 3900–3913. [CrossRef] [PubMed] Molecules 2019, 24, 837 17 of 21 14. Locatelli, C.; Rosso, R.; Santos-Silva, M.C.; de Souza, C.A.; Licínio, M.A.; Leal, P.; Bazzo, M.L.; Yunes, R.A.; Creczynski-Pasa, T.B. Ester derivatives of gallic acid with potential toxicity toward L1210 leukemia cells. Bioorg. Med. Chem. 2008, 16, 3791–3799. [CrossRef] [PubMed] 15. Fernandes, I.; Pérez-Gregorio, R.; Soares, S.; Mateus, N.; de Freitas, V. Wine Flavonoids in Health and Disease Prevention. Molecules 2017, 22, 292. [CrossRef] [PubMed] 16. Fiuza, S.M.; Gomes, C.; Teixeira, L.J.; Girão da Cruz, M.T.; Cordeiro, M.N.D.S.; Milhazes, N.; Borges, F.; Marques, M.P.M. Phenolic acid derivatives with potential anticancer properties—A structure-activity relationship study. Part 1: Methyl. propyl and octyl esters of caffeic and gallic acids. Bioorg. Med. Chem. 2004, 12, 3581–3589. [CrossRef] [PubMed] 17. Sergediene, ˙ E.; Jönsson, K.; Szymusiak, H.; Tyrakowska, B.; Rietjens, I.M.C.M.; Cenas, ˙ N. Prooxidant toxicity of polyphenolic antioxidants to HL-60 cells: Description of quantitative structure-activity relationships. FEBS Lett. 1999, 462, 392–396. [CrossRef] ˇ ˇ 18. Nemeikaite- ˙ Cenien ˙ e, ˙ A.; Imbrasaite, ˙ A.; Sergediene, ˙ E.; Cenas, ˙ N. Quantitative structure-activity relationships in prooxidant cytotoxicity of polyphenols: Role of potential of phenoxyl radical/phenol redox couple. Archives Biochem. Biophys. 2005, 441, 182–190. [CrossRef] [PubMed] 19. Gomes, C.A.; da Cruz, T.G.; Andrade, J.L.; Milhazes, N.; Borges, F.; Marques, M.P.M. Anticancer activity of phenolic acids of natural or synthetic origin: A structure-activity study. J. Med. Chem. 2003, 46, 5395–5401. [CrossRef] [PubMed] 20. Nagaoka, T.; Banskota, A.H.; Tezuka, Y.; Saiki, I.; Kadota, S. Selective antiproliferative activity of caffeic acid phenethyl aster analogues on highly liver-metastatic murine colon 26-L5 carcinoma cell line. Bioorg. Med. Chem. 2002, 10, 3351–3359. [CrossRef] 21. Blank, V.C.; Poli, C.; Marder, M.; Roguin, L.P. 2004 Antiproliferative activity of various ﬂavonoids and related compounds: Additive effect of interferon-2b. Bioorg. Med. Chem. Lett. 2004, 14, 133–136. [CrossRef] [PubMed] 22. Chun, K.; Park, S.K.; Kim, H.M.; Choi, Y.; Kim, M.H.; Park, C.H.; Joe, B.Y.; Chun, T.G.; Choi, H.M.; Lee, H.Y.; et al. Chromen-based TNF- converting enzyme (TACE) inhibitors: Design synthesis and biological evaluation. Bioorg. Med. Chem. 2008, 16, 530–535. [CrossRef] [PubMed] 23. Lee, Y.T.; Don, M.J.; Liao, C.H.; Chiou, H.W.; Chen, C.F.; Ho, L.K. Effects of phenolic acid esters and amides on stimulus-induced reactive oxygen species production in human neutrophils. Clin. Chim. Acta 2005, 352, 135–141. [CrossRef] [PubMed] 24. Rao, Y.K.; Fang, S.H.; Tzeng, Y.M. Differential effects of synthesized 2 -oxygenated chalcone derivatives: Modulation of human cell cycle phase distribution. Bioorg. Med. Chem. 2004, 12, 2679–2686. [CrossRef] [PubMed] 25. Rao, K.Y.; Fang, S.H.; Tzeng, Y.M. Synthesis, growth inhibition, and cell cycle evaluations of novel ﬂavonoid derivatives. Bioorg. Med. Chem. 2005, 13, 6850–6855. [CrossRef] [PubMed] 26. Fang, S.H.; Rao, Y.K.; Tzeng, Y.M. Inhibitory effects of ﬂavonol glycosides from Cinnamomum osmophloeum on inﬂammatory mediators in LPS/IFN- -activated murine macrophages. Bioorg. Med. Chem. 2005, 13, 2381–2388. [CrossRef] [PubMed] 27. Benavente-García, O.; Castillo, J. Update on uses and properties of citrus ﬂavonoids: New ﬁndings in anticancer, cardiovascular, and anti-inﬂammatory activity. J. Agric. Food Chem. 2008, 56, 6185–6205. [CrossRef] [PubMed] 28. Cardenas, M.; Marder, M.; Blank, V.C.; Roguin, L.P. Antitumor activity of some natural ﬂavonoids and derivatives on various human and maurine cancer cell lines. Bioorg. Med. Chem. 2006, 14, 2966–2971. [CrossRef] [PubMed] 29. Bendini, A.; Cerretani, L.; Carrasco-Pancorbo, A.; Gómez-Caravaca, A.M.; Segura-Carretero, A.; Fernández-Gutiérrez, A.; Lercker, G. Phenolic Molecules in Virgin Olive Oils: A Survey of Their Sensory Properties, Health Effects, Antioxidant Activity and Analytical Methods. An Overview of the Last Decade. Molecules 2007, 12, 1679–1719. [CrossRef] [PubMed] 30. Seraﬁm, T.L.; Carvalho, F.S.; Maeques, M.P.M.; Chaleiros, R.; Silva, T.; Garrido, J.; Milhazes, N.; Borges, F.; Roleira, F.; Silva, E.T.; et al. Lipophilic caffeic and ferulic acid derivatives presenting cytotoxicity against human breast cancer cells. Chem. Res. Toxicol. 2011, 24, 763–774. [CrossRef] [PubMed] Molecules 2019, 24, 837 18 of 21 31. Li, F.; Wu, J.-J.; Wang, J.; Yang, X.-L.; Cai, P.; Liu, Q.-H.; Kong, L.-Y.; Wang, X.-B. Synthesis and pharmacological evaluation of novel chromone derivatives as balanced multifunctional agents against Alzheimer ’s disease. Bioorg. Med. Chem. 2017, 25, 3815–3826. [CrossRef] [PubMed] 32. Marsh, D.T.; Das, S.; Ridell, J.; Smid, S.D. Structure-activity relationships for ﬂavone interactions with 0 0 0 amyloid reveal a novel anti-aggregatory and neuroprotective effect of 2 ,3 ,4 -trihydroxyﬂavone (2-D08). Bioorg. Med. Chem. 2017, 25, 3827–3834. [CrossRef] [PubMed] 33. Letenneur, L.; Proust-Lima, C.; Le Gouge, A.; Dartigues, J.F.; Barberger-Gateau, P. Flavonoid intake and cognitive decline over a 10-year period. Am. J. Epidemiol. 2007, 165, 1364–1371. [CrossRef] [PubMed] 34. Manach, C.; Williamson, G.; Morand, C.; Scalbert, A.; Ramesy, C. Bioavailability and bioefﬁcacy of polyphenols in humans. Review of 97 bioavailability studies. Am. J. Clin. Nutr. 2005, 81, 230S–242S. [CrossRef] [PubMed] 35. Passamonti, S.; Terdoslavich, M.; Franca, R.; Vanzo, A.; Tramer, F.; Braidot, E.; Petrussa, E.; Vianello, A. Bioavailability of ﬂavonoids: A review of their membrane transport and the function of bilitranslocase in animal and plant organisms. Curr. Drug Metabol. 2009, 10, 369–394. [CrossRef] 36. Mennen, L.I.; Sapinho, D.; Ito, H.; Galan, P.; Hercberg, S.; Scalbert, A. Urinary excretion of 13 dietary ﬂavonoids and phenolic acids in free-living healthy subjects - variability and possible use as biomarkers of polyphenol intake. Eur. J. Clin. Nutr. 2008, 62, 519–525. [CrossRef] [PubMed] 37. Xie, H. Activity assay of membrane transport proteins. Acta Biochim. Biophys. Sin. 2008, 40, 269–277. [CrossRef] [PubMed] 38. Billat, P.A.; Roger, A.; Faure, S.; Lagarce, F. Models for drug absorption from the small intestine: Where are we and where are we going? Drug Discovery Today 2017, 22, 761–775. [CrossRef] [PubMed] 39. De Bruyn, T.; van Westen, G.J.P.; Jzerman, A.P.I.; Stieger, B.; de Witte, P.; Augustijns, P.F.; Annaert, P.P. Structure-Based Identiﬁcation of OATP1B1/3 Inhibitors. Mol. Pharmacol. 2013, 83, 1257–1267. [CrossRef] [PubMed] 40. Venko, K.; Roy Choudhury, A.; Novic, ˇ M. Computational Approaches for Revealing the Structure of Membrane Transporters: Case Study on Bilitranslocase. Comput. Struct. Biotech. J. 2017, 15, 232–242. [CrossRef] [PubMed] 41. Liang, J.; Naveed, H.; Jimenez-Morales, D.; Adamian, L.; Lin, M. Computational studies of membrane proteins: Models and predictions for biological understanding. Biochim. Biophys. Acta 2012, 1818, 927–941. [CrossRef] [PubMed] 42. Russel, D.; Lasker, K.; Webb, B.; Velázquez-Muriel, J.; Tjioe, E.; Schneidman-Duhovny, D.; Peterson, B.; Sali, A. Putting the pieces together: Integrative structure determination of macromolecular assemblies. PLoS Biol. 2012, 10, e1001244. [CrossRef] [PubMed] 43. Xu, Y.; Shen, Q.; Liu, X.; Lu, J.; Li, S.; Luo, C.; Gong, L.; Luo, X.; Zheng, M.; Jiang, H. Computational models for predicting interactions with membrane transporters. Curr. Med. Chem. 2013, 20, 2118–2136. [CrossRef] [PubMed] 44. Sedykh, A.; Fourches, D.; Duan, J.; Hucke, O.; Garneau, M.; Zhu, H.; Bonneau, P.; Tropsha, A. Human intestinal transporter database: QSAR modeling and virtual proﬁling of drug uptake, efﬂux and interactions. Pharm Res. 2013, 30, 996–1007. [CrossRef] [PubMed] 45. Chang, C.; Ray, A.; Swaan, P. In silico strategies for modeling membrane transporter function. Drug Discovery Today 2005, 10, 663–671. [CrossRef] 46. Mak, L.; Marcus, D.; Howlett, A.; Yarova, G.; Duchateau, G.; Klake, W.; Bender, A.; Glen, R.C. Metrabase: A cheminformatics and bioinformatics database for small molecule transporter data analysis and (Q)SAR modeling. J. Cheminform. 2015, 7, 31. [CrossRef] [PubMed] 47. Elbourne, L.D.H.; Tetu, S.; Hassan, K.; Paulsen, I.T. TransportDB 2.0: A database for exploring membrane transporters in sequenced genomes from all domains of life. Nucleic Acids Res. 2017, 45, D320–D324. [CrossRef] [PubMed] 48. Saier, M.H.; Reddy, V.S.; Tsu, B.V.; Ahmed, M.S.; Li, C.; Moreno-Hagelsieb, G. (The Transporter Classiﬁcation Database (TCDB): Recent advances. Nucleic Acids Res. 2016, 44, D372–D379. [CrossRef] [PubMed] 49. Battiston, L.; Passamonti, S.; Macagno, A.; Sottocasa, G.L. The bilirubin-binding motif of bilitranslocase and its relation to conserved motifs in ancient biliproteins. Biochem. Biophys. Res. Commun. 1998, 247, 687–692. [CrossRef] [PubMed] Molecules 2019, 24, 837 19 of 21 50. Niemi, M.; Pasanen, M.K.; Neuvonen, P.J. Organic anion transporting polypeptide 1B1: A genetically polymorphic transporter of major importance for hepatic drug uptake. Pharmacol. Rev. 2011, 63, 157–181. [CrossRef] [PubMed] 51. Passamonti, S.; Terdoslavich, M.; Margon, A.; Cocolo, A.; Medic, N.; Micali, F.; Decorti, G.; Franko, M. Uptake of bilirubin into HepG2 cells assayed by thermal lens spectroscopy. Function of bilitranslocase. FEBS J. 2005, 272, 5522–5535. [CrossRef] [PubMed] 52. Cvorovic, ´ J.; Passamonti, S. Membrane transporters for bilirubin and its conjugates: A systematic review. Front. Pharmacol. 2017, 8, 887. [CrossRef] [PubMed] 53. Karawajczyk, A.; Drgan, V.; Medic, N.; Oboh, G.; Passamonti, S.; Novic, ˇ M. Properties of ﬂavonoids inﬂuencing the binding to bilitranslocase investigated by neural network modelling. Biochem. Pharm. 2007, 73, 308–320. [CrossRef] [PubMed] 54. Župerl, Š.; Fornasaro, S.; Novic, ˇ M.; Passamonti, S. Experimental determination and prediction of bilitranslocase transport activity. Anal. Chim. Acta 2011, 705, 322–333. [CrossRef] [PubMed] 55. Maestro, A.; Terdoslavich, M.; Vanzo, A.; Kuku, A.; Tramer, F.; Nicolin, V.; Micali, F.; Decorti, G.; Passamonti, S. Expression of bilitranslocase in the vascular endothelium and its function as a ﬂavonoid transporter. Cardiovasc. Res. 2010, 85, 175–183. [CrossRef] [PubMed] 56. Nicolin, V.; Grill, V.; Micali, F.; Narducci, P.; Passamonti, S. Immunolocalisation of bilitranslocase in mucosecretory and parietal cells of the rat gastric mucosa. J. Mol. Histol. 2005, 36, 45–50. [CrossRef] [PubMed] 57. Brandoni, A.; Di Giusto, G.; Franca, R.; Passamonti, S.; Torres, A.M. Expression of kidney and liver bilitranslocase in response to acute biliary obstruction. Nephron. Physiol. 2010, 114, 35–40. [CrossRef] [PubMed] 58. Cvorovic, ´ J.; Ziberna, L.; Fornasaro, S.; Tramer, F.; Passamonti, S. Bioavailability of Flavonoids: The Role of Cell Membrane Transporters. In Polyphenols: Mechanisms of Action in Human Health and Disease, 2nd ed.; Watson, R.R., Preedy, V.R., Zibadi, S., Eds.; Elsevier Academic Press: Amsterdam, The Netherlands, 2018; pp. 295–320. 59. Martinci ˇ c, ˇ R.; Venko, K.; Župerl, Š.; Novic, ˇ M. Chemometrics approach for the prediction of structure-activity relationship for membrane transporter bilitranslocase. SAR QSAR Env. Res. 2014, 25, 853–872. [CrossRef] [PubMed] 60. Montanic, S.; Terdoslavich, M.; Rajcevic, U.; De Leo, L.; Bonin, S.; Curin Serbec, V.; Passamonti, S. Development and characterization of a novel mAb against bilitranslocase—A new biomarker of renal carcinoma. Radiol. Oncol. 2013, 47, 128–137. [CrossRef] [PubMed] 61. Golijanin, D.J.; Marshall, J.; Cardin, A.; Singer, E.A.; Wood, R.W.; Reeder, J.E.; Wu, G.; Yao, J.L.; Passamonti, S.; Messing, E.M. Bilitranslocase (BTL) is immunolocalised in proximal and distal renal tubules and absent in renal cortical tumors accurately corresponding to intraoperative near infrared ﬂuorescence (NIRF) expression of renal cortical tumors using intravenous indocyanine green (ICG). J. Urol. 2008, 179, 137. 62. Roy Choudhury, A.; Sikorska, E.; van den Boom, J.; Bayer, P.; Popenda, L.; Szutkowski, K.; Jurga, S.; Bonomi, M.; Sali, A.; Zhukov, I.; et al. Structural Model of the Bilitranslocase Transmembrane Domain Supported by NMR and FRET Data. PLoS ONE 2015, 10, e0135455. [CrossRef] [PubMed] 63. Passamonti, S.; Vrhovsek, U.; Vanzo, A.; Mattivi, F. The stomach as a site for anthocyanins absorption from food. FEBS Lett. 2003, 544, 210–213. [CrossRef] 64. Chirico, N.; Gramatica, P. Real External Predictivity of QSAR Models. Part 2. New Intercomparable Thresholds for Different Validation Criteria and the Need for Scatter Plot Inspection. J. Chem. Inf. Model. 2012, 52, 2044–2058. [CrossRef] [PubMed] 65. Todeschini, R.; Consonni, V.; Gramatica, P. Chemometrics in QSAR. In Comprehensive Chemometrics, Volume 4; Brown, S.D., Tauler, R., Walczak, B., Eds.; Elsevier: Oxford, UK, 2009; pp. 129–172. 66. Ruiz, I.L.; Gómez-Nieto, M.A. Study of the Applicability Domain of the QSAR Classiﬁcation Models by Means of the Rivality and Modelability Indexes. Molecules 2018, 23, 2756. [CrossRef] [PubMed] 67. Minovski, N.; Župerl, Š.; Drgan, V.; Novic, ˇ M. Assessment of applicability domain for multivariate counter-propagation artiﬁcial neural network predictive models by minimum Euclidean distance space analysis: A case study. Anal. Chim. Acta 2013, 759, 28–42. [CrossRef] [PubMed] 68. Gramatica, P.; Chirico, N.; Papa, E.; Kovarich, S.; Cassani, S. QSARINS: A New Software for the Development, Analysis, and Validation of QSAR MLR Models. J. Comput. Chem. 2013, 34, 2121–2132. [CrossRef] Molecules 2019, 24, 837 20 of 21 69. Marzo, M.; Kulkarni, S.; Manganaro, A.; Roncaglioni, A.; Wu, S.; Barton-Maclaren, T.S.; Lester, C.; Benfenati, E. Integrating in silico models to enhance predictivity for developmental toxicity. Toxicology 2016, 370, 127–137. [CrossRef] [PubMed] 70. Wang, X.; Wang, Q.; Morris, M.E. Pharmacokinetic interaction between the ﬂavonoid luteolin and gamma-hydoxybutyrate in rats: Potential involvement of monocarboxylate transporters. APPS J. 2008, 10, 47–55. 71. Piskula, M.K.; Yamakoshi, J.; Iwai, Y. Daizdein and genistein but not their glucosides are absorbed from the rat stomach. FEBS Lett. 1999, 447, 287–291. [CrossRef] 72. Roy, P.P.; Paul, S.; Mitra, I.; Roy, K. On two novel parameters for validation of predictive QSAR models. Molecules 2009, 14, 1660–1701. 73. Rodriguez-Morato, J.; Xicota, L.; Fito, M.; Farre, M.; Dierssen, M.; de la Torre, R. Potential role of olive oil phenolic compounds in the prevention of neurodegenerative diseases. Molecules 2015, 20, 4655–4680. [CrossRef] [PubMed] 74. Mathew, B.C.; Yoseph, B.A.; Dessale, T.; Daniel, R.S.; Alemayehu, A.; Campbell, I.W.; Augusti, K.T. Hypolipidaemic effect of leucodelphinidin derivative from Ficus bengalensis (Linn) on cholesterol fed rats. Res. J. Chem. Sci. 2012, 2, 54–60. 75. Terdoslavich, M.; de Graaf, I.A.M.; Proost, J.H.; Cocolo, A.; Passamonti, S.; Groothuis, G.M.M. Bilitranslocase is Involved in the Uptake of Bromosulfophthalein in Rat and Human Liver. Drug Metab. Lett. 2012, 6, 165–173. [CrossRef] [PubMed] 76. Bielen, L.; Kralj, I.; Curci ˇ c, ´ E.; Vodanovic, ´ M.; Boban, A.; Božina, N. Acute kidney injury, agranulocytosis, drug-induced liver injury, and posterior reversible encephalopathy syndrome caused by high-dose methotrexate—Possible role of low activity ABC and SLC drug transporters. Eu. J. Clinic. Pharmacol. 2018, 74, 1191–1192. [CrossRef] [PubMed] 77. Wilson, A.G.E. New Horizons in Predictive Drug Metabolism and Pharmacokinetics; Royal Society of Chemistry: Cambridge, UK, 2015; p. 422. 78. Zhang, G.; Ren, Y. Molecular Modeling and Design Studies of Purine Derivatives as Novel CDK2 Inhibitors. Molecules 2018, 23, 2924. [CrossRef] [PubMed] 79. Matyugina, E.; Khandazhinskaya, A.; Chernousova, L.; Andreevskaya, S.; Smirnova, T.; Chizhov, A.; Karpenko, I.; Kochetkov, S.; Alexandrova, L. The synthesis and antituberculosis activity of 50-nor carbocyclic uracil derivatives. Bioorg. Med. Chem. 2012, 20, 6680–6686. [CrossRef] [PubMed] 80. Okazaki, S.; Mizuhara, T.; Shimura, K.; Murayama, H.; Ohno, H.; Oishi, S.; Matsuoka, M.; Fujii, N. Identiﬁcation of anti-HIV agents with a novel benzo [4,5]isothiazolo[2,3-a]pyrimidine scaffold. Bioorg. Med. Chem. 2015, 23, 1447–1452. [CrossRef] [PubMed] 81. Sosnin, S.; Vashurina, M.; Withnall, M.; Karpov, P.; Fedorov, M.; Tetko, I.V. A Survey of Multi-Task Learning Methods in Chemoinformatics. Mol. Inf. 2018, 37, 1800108. [CrossRef] [PubMed] 82. Simões, R.S.; Maltarollo, V.G.; Oliveira, P.R.; Honorio, K.M. Transfer and Multi-task Learning in QSAR Modeling: Advances and Challenges. Front. Pharmacol. 2018, 9, 74. [CrossRef] [PubMed] 83. Žitnik, M.; Zupan, B. Matrix factorization-based data fusion for drug-induced liver injury prediction. Systems Biomed. 2014, 2, 16–22. [CrossRef] 84. Worth, A.P. The future of in silico chemical safety . . . and beyond. Comput. Toxicol. 2019, 10, 60–62. [CrossRef] 85. Dias, M.M.; Machado, N.F.L.; Marques, M.P.M. Dietary chromomes as antioxidant agents—the structural variable. Food Funct. 2011, 2, 595–602. [CrossRef] [PubMed] 86. Paiva-Martins, F.; Rodrigues, V.; Chalheiros, R.; Marques, M.P.M. Characterization of antioxidant olive oil biophenols by spectroscopic methods. J. Sci. Food Agric. 2011, 91, 309–314. [CrossRef] [PubMed] 87. Pisani, L.; Farina, R.; Nicolotti, O.; Gadaleta, D.; Soto-Otero, R.; Catto, M.; Di Braccio, M.; Mendez-Alvarez, E.; Carotti, A. In silico design of novel 2H-chromen-2-one derivatives as potent and selective MAO-B inhibitors. Eur. J. Med. Chem. 2015, 89, 98–105. [CrossRef] [PubMed] 88. Yildiz, S.Z.; Küçükislamoglu, ˘ M.; Tuna, M. Synthesis and characterization of novel ﬂavonoid-substituted phthalocyanines using ()naringenin. J. Organometal. Chem. 2009, 694, 4152–4161. [CrossRef] 89. Cao, G.; Soﬁc, E.; Prior, R.L. Antioxidant and prooxidant behavior of ﬂavonoids: Structure-activity relationships. Free Radical Biol. Med. 1997, 22, 749–760. [CrossRef] Molecules 2019, 24, 837 21 of 21 90. Petrussa, E.; Braidot, E.; Zancani, M.; Peresson, C.; Bertolini, A.; Patui, S.; Vianello, A. Plant Flavonoids—Biosynthesis, Transport and Involvement in Stress Responses. Int. J. Mol. Sci. 2013, 14, 14950–14973. [CrossRef] [PubMed] 91. Silva, F.A.M.; Borges, F.; Guimarães, C.; Lima, J.L.F.C.; Matos, C.; Reis, S. Phenolic acids and derivatives: Studies on the relationship among structure, radical scavenging activity, and physicochemical parameters. J. Agric. Food Chem. 2000, 48, 2122–2126. [CrossRef] [PubMed] 92. Heim, K.E.; Tagliaferro, A.R.; Bobilya, D.J. Flavonoid antioxidants: Chemistry, metabolism and structure-activity relationships. J. Nutr. Biochem. 2002, 13, 572–582. [CrossRef] 93. Martinci ˇ c, ˇ R.; Mravljak, J.; Švajger, U.; Perdih, A.; Andreluh, M.; Novic, ˇ M. In silico discovery of novel potent antioxidants on the basis of pulvinic acid and coumarin derivatives and their experimental evaluation. PLoS ONE 2015, 10, e0140602. [CrossRef] [PubMed] 94. Dewar, M.J.S.; Zoebisch, E.G.; Healy, E.F.; Stewart, J.J.P. Development and use of quantum mechanical molecular models. AM1: A new general purpose quantum mechanical molecular model. J. Am. Chem. Soc. 1985, 107, 3902–3909. 95. Katritzky, A.R.; Lobanov, V.S.; Karelson, M. Codessa 2.0. Comprehensive Descriptors for Structural and Statistical Analysis; University of Florida: Gainesville, FL, USA, 1994. 96. Dragon 7.0—Software for Molecular Descriptors Calculation. 2016. Available online: https://chm.kode-solutions. net/products_dragon.php (accessed on 20 August 2018). 97. Drgan, V.; Župerl, Š.; Vracko, ˇ M.; Cappelli, C.I.; Novic, ˇ M. CPANNatNIC software for counter-propagation neural network to assist in read-across. J. Cheminform. 2017, 9, 30. [CrossRef] [PubMed] 98. Drgan, V.; Župerl, Š.; Vracko, ˇ M.; Como, F.; Novic, ˇ M. Robust modelling of acute toxicity towards fathead minnow (Pimephales promelas) using counter-propagation artiﬁcial neural networks and genetic algorithm. SAR QSAR Environ. Res. 2016, 27, 501–519. [CrossRef] [PubMed] 99. Boughorbel, S.; Jarray, F.; El-Anbari, M. Optimal classiﬁer for imbalanced data using Matthews Correlation Coefﬁcient metric. PLoS ONE 2017, 12, e0177678. [CrossRef] [PubMed] 100. Fawcett, T. An introduction to ROC analysis. Pattern Recognition Lett. 2006, 27, 861–874. [CrossRef] 101. Metrabase. Available online: www-metrabase.ch.cam.ac.uk (accessed on 20 November 2018). 102. Chembench. Available online: https://chembench.mml.unc.edu (accessed on 4 December 2018). Models Availability: Newly developed classiﬁcation models for the prediction of bilitranslocase transport activity are available upon request from the authors. © 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Journal

Molecules – Multidisciplinary Digital Publishing Institute

Published: Feb 27, 2019

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase

An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase

An In Silico Approach for Assessment of the Membrane Transporter Activities of Phenols: A Case Study Based on Computational Models of Transport Activity for the Transporter Bilitranslocase

References (108)

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies