Prediction of cytochrome P450 3A4, 2D6, and 2C9 inhibitors and substrates by using support vector machines

Advances and applications of machine learning and deep learning in environmental ecology and health

Machine learning (ML) and deep learning (DL) possess excellent advantages in data analysis (e.g., feature extraction, clustering, classification, regression, image recognition and prediction) and risk assessment and management in environmental ecology and health (EEH). Considering the rapid growth and increasing complexity of data in EEH, it is of significance to summarize recent advances and applications of ML and DL in EEH. This review summarized the basic processes and fundamental algorithms of the ML and DL modeling, and indicated the urgent needs of ML and DL in EEH. Recent research hotspots such as environmental ecology and restoration, environmental fate of new pollutants, chemical exposures and risks, chemical hazard identification and control were highlighted. Various applications of ML and DL in EEH demonstrate their versatility and technological revolution, and present some challenges. The perspective of ML and DL in EEH were further outlined to promote the innovative analysis and cultivation of the ML-driven research paradigm.

Machine Learning Models to Predict Cytochrome P450 2B6 Inhibitors and Substrates

Cytochrome P450 2B6 (CYP2B6) is responsible for the metabolism of ∼7% of marketed drugs. The in vitro drug interaction studies guidance for industry issued by the FDA stipulates that drug sponsors need to evaluate whether the investigated drugs interact with the major drug-metabolizing P450s including CYP2B6. Therefore, there has been greater attention to the development of predictive models for CYP2B6 inhibitors and substrates. In this study, conventional machine learning and deep learning models were developed to predict CYP2B6 inhibitors and substrates. Our results showed that the best CYP2B6 inhibitor model yielded the AUC values of 0.95 and 0.75 with the 10-fold cross-validation and the test set, respectively, and the best CYP2B6 substrate model produced the AUC values of 0.93 and 0.90 with the 10-fold cross-validation and the test set, respectively. The generalization ability of the CYP2B6 inhibitor and substrate models was assessed by using the external validation sets. Several significant substructural fragments relevant to CYP2B6 inhibitors and substrates were detected via frequency substructure analysis and information gain. In addition, the applicability domain of the models was defined by employing a nonparametric method based on the probability density distribution. We anticipate that our results would be useful for the prediction of potential CYP2B6 inhibitors and substrates in the early stage of drug discovery.

3D-QSAR Studies, Molecular Docking, Molecular Dynamic Simulation, and ADMET Proprieties of Novel Pteridinone Derivatives as PLK1 Inhibitors for the Treatment of Prostate Cancer

Overexpression of polo-like kinase 1 (PLK1) has been found in many different types of cancers. With its essential role in cell proliferation, PLK1 has been determined to be a broad-spectrum anti-cancer target. In this study, 3D-QSAR, molecular docking, and molecular dynamics (MD) simulations were applied on a series of novel pteridinone derivatives as PLK1 inhibitors to discover anti-cancer drug candidates. In this work, three models—CoMFA (Q² = 0.67, R² = 0.992), CoMSIA/SHE (Q² = 0.69, R² = 0.974), and CoMSIA/SEAH (Q² = 0.66, R² = 0.975)—of pteridinone derivatives were established. The three models that were established gave Rpred2 = 0.683, Rpred 2= 0.758, and Rpred 2= 0.767, respectively. Thus, the predictive abilities of the three proposed models were successfully evaluated. The relations between the different champs and activities were well-demonstrated by the contour chart of the CoMFA and CoMSIA/SEAH models. The results of molecular docking indicated that residues R136, R57, Y133, L69, L82, and Y139 were the active sites of the PLK1 protein (PDB code: 2RKU), in which the more active ligands can inhibit the enzyme of PLK1. The results of the molecular dynamic MD simulation diagram were obtained to reinforce the previous molecular docking results, which showed that both inhibitors remained stable in the active sites of the PLK1 protein (PDB code: 2RKU) for 50 ns. Finally, a check of the ADME-Tox properties of the two most active molecules showed that molecular N° 28 could represent a good drug candidate for the therapy of prostate cancer diseases.

Machine learning to predict metabolic drug interactions related to cytochrome P450 isozymes

Drug–drug interaction (DDI) often causes serious adverse reactions and thus results in inestimable economic and social loss. Currently, comprehensive DDI evaluation has become a major challenge in pharmaceutical research due to the time-consuming and costly process of the experimental assessment and it is of high necessity to develop effective in silico methods to predict and evaluate DDIs accurately and efficiently. In this study, based on a large number of substrates and inhibitors related to five important CYP450 isozymes (CYP1A2, CYP2C9, CYP2C19, CYP2D6 and CYP3A4), a series of high-performance predictive models for metabolic DDIs were constructed by two machine learning methods (random forest and XGBoost) and 4 different types of descriptors (MOE_2D, CATS, ECFP4 and MACCS). To reduce the uncertainty of individual models, the consensus method was applied to yield more reliable predictions. A series of evaluations illustrated that the consensus models were more reliable and robust for the DDI predictions of new drug combination. For the internal validation, the whole prediction accuracy and AUC value of the DDI models were around 0.8 and 0.9, respectively. When it was applied to the external datasets, the model accuracy was 0.793 and 0.795 for multi-level validation and external validation, respectively. Furthermore, we also compared our model with some recently published tools and then applied the final model to predict FDA-approved drugs and proposed 54,013 possible drug pairs with potential DDIs. In summary, we developed a powerful DDI predictive model from the perspective of the CYP450 enzyme family and it will help a lot in the future drug development and clinical pharmacy research.

The FDA-Approved Drug Cobicistat Synergizes with Remdesivir To Inhibit SARS-CoV-2 Replication In Vitro and Decreases Viral Titers and Disease Progression in Syrian Hamsters

Combinations of direct-acting antivirals are needed to minimize drug resistance mutations and stably suppress replication of RNA viruses. Currently, there are limited therapeutic options against the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), and testing of a number of drug regimens has led to conflicting results. Here, we show that cobicistat, which is an FDA-approved drug booster that blocks the activity of the drug-metabolizing proteins cytochrome P450-3As (CYP3As) and P-glycoprotein (P-gp), inhibits SARS-CoV-2 replication. Two independent cell-to-cell membrane fusion assays showed that the antiviral effect of cobicistat is exerted through inhibition of spike protein-mediated membrane fusion. In line with this, incubation with low-micromolar concentrations of cobicistat decreased viral replication in three different cell lines including cells of lung and gut origin. When cobicistat was used in combination with remdesivir, a synergistic effect on the inhibition of viral replication was observed in cell lines and in a primary human colon organoid. This was consistent with the effects of cobicistat on two of its known targets, CYP3A4 and P-gp, the silencing of which boosted the in vitro antiviral activity of remdesivir in a cobicistat-like manner. When administered in vivo to Syrian hamsters at a high dose, cobicistat decreased viral load and mitigated clinical progression. These data highlight cobicistat as a therapeutic candidate for treating SARS-CoV-2 infection and as a potential building block of combination therapies for COVID-19.

DeepStack-DTIs: Predicting Drug-Target Interactions Using LightGBM Feature Selection and Deep-Stacked Ensemble Classifier

Accurate prediction of drug-target interactions (DTIs), which is often used in the fields of drug discovery and drug repositioning, is regarded a key challenge in the study of drug science. In this paper, a new method called DeepStack-DTIs is proposed to predict DTIs. First, for the target protein, pseudo-position specific score matrix, pseudo amino acid composition and SPIDER3 are used to extract the different feature information of the target protein. Meanwhile, the path-based fingerprint features of each drug are extracted. Then, the synthetic minority oversampling technique (SMOTE) and light gradient boosting machine (LightGBM) are used for data balancing and feature selection, respectively. Finally, the processed features are input to the deep-stacked ensemble classifier composed of gated recurrent unit (GRU), deep neural network (DNN), support vector machine (SVM), eXtreme gradient boosting (XGBoost) and logistic regression (LR) to predict DTIs. Under the five-fold cross-validation and compared with existing methods, the proposed method achieves higher prediction accuracy on the gold standard dataset. To evaluate the predictive power of DeepStack-DTIs, we validate the method on another dataset and predict the drug-target interaction network. The results indicate that DeepStack-DTIs has excellent predictive ability than the other methods, and provides novel insights for the prediction of DTIs. A novel method DeepStack-DTIs for drug-target interactions prediction. PsePSSM, PseAAC, SPIDER3 and FP2 are fused to convert protein sequence and drug molecule information into digital information, respectively. The SMOTE algorithm is used to balance the dataset and LightGBM feature selection algorithm is employed to remove redundant and irrelevant features to select the optimal feature subset. This optimal feature subset is inputted into the deep-stacked ensemble classifier to predict drug-target interactions. The experimental results show DeepStack-DTIs method can significantly improve the prediction accuracy of drug-target interactions.

ADMET Predictability at Boehringer Ingelheim: State-of-the-Art, and Do Bigger Datasets or Algorithms Make a Difference?

Computational methods assisting drug discovery and development are routine in the pharmaceutical industry. Digital recording of ADMET assays has provided a rich source of data for development of predictive models. Despite the accumulation of data and the public availability of advanced modeling algorithms, the utility of prediction in ADMET research is not clear. Here, we present a critical evaluation of the relationships between data volume, modeling algorithm, chemical representation and grouping, and temporal aspect (time sequence of assays) using an in-house ADMET database. We find no large difference in prediction algorithms nor any systemic and substantial gain from increasingly large datasets. Temporal-based data enlargement led to performance improvement in only in a limited number of assays, and with fractional improvement at best. Assays that are well-, intermediately-, or poorly-suited for ADMET predictions and reasons for such behavior are systematically identified, generating realistic expectations for areas in which computational models can be used to guide decision making in molecular design and development.

Molecular modeling approaches to address drug-metabolizing enzymes (DMEs) mediated chemoresistance: a review

Resistance against clinically approved anticancer drugs is the main roadblock in cancer treatment. Drug metabolizing enzymes (DMEs) that are capable of metabolizing a variety of xenobiotic get overexpressed in malignant cells, therefore, catalyzing drug inactivation. As evident from the literature reports, the levels of DMEs increase in cancer cells that ultimately lead to drug inactivation followed by drug resistance. To puzzle out this issue, several strategies inclusive of analog designing, prodrug designing, and inhibitor designing have been forged. On that front, the implementation of computational tools can be considered a fascinating approach to address the problem of chemoresistance. Various research groups have adopted different molecular modeling tools for the investigation of DMEs mediated toxicity problems. However, the utilization of these in-silico tools in maneuvering the DME mediated chemoresistance is least considered and yet to be explored. These tools can be employed in the designing of such chemotherapeutic agents that are devoid of the resistance problem. The current review canvasses various molecular modeling approaches that can be implemented to address this issue. Special focus was laid on the development of specific inhibitors of DMEs. Additionally, the strategies to bypass the DMEs mediated drug metabolism were also contemplated in this report that includes analogs and pro-drugs designing. Different strategies discussed in the review will be beneficial in designing novel chemotherapeutic agents that depreciate the resistance problem.

In-Silico Modeling in Drug Metabolism and Interaction: Current Strategies of Lead Discovery

BACKGROUND

Drug metabolism is a complex mechanism of human body systems to detoxify foreign particles, chemicals, and drugs through bio alterations. It involves many biochemical reactions carried out by invivo enzyme systems present in the liver, kidney, intestine, lungs, and plasma. After drug administration, it crosses several biological membranes to reach into the target site for binding and produces the therapeutic response. After that, it may undergo detoxification and excretion to get rid of the biological systems. Most of the drugs and its metabolites are excreted through kidney via urination. Some drugs and their metabolites enter into intestinal mucosa and excrete through feces. Few of the drugs enter into hepatic circulation where they go into the intestinal tract. The drug leaves the liver via the bile duct and is excreted through feces. Therefore, the study of total methodology of drug biotransformation and interactions with various targets is costly.

METHODS

To minimize time and cost, in-silico algorithms have been utilized for lead-like drug discovery. Insilico modeling is the process where a computer model with a suitable algorithm is developed to perform a controlled experiment. It involves the combination of both in-vivo and in-vitro experimentation with virtual trials, eliminating the non-significant variables from a large number of variable parameters. Whereas, the major challenge for the experimenter is the selection and validation of the preferred model, as well as precise simulation in real physiological status.

RESULTS

The present review discussed the application of in-silico models to predict absorption, distribution, metabolism, and excretion (ADME) properties of drug molecules and also access the net rate of metabolism of a compound.

CONCLUSION

It helps with the identification of enzyme isoforms; which are likely to metabolize a compound, as well as the concentration dependence of metabolism and the identification of expected metabolites. In terms of drug-drug interactions (DDIs), models have been described for the inhibition of metabolism of one compound by another, and for the compound-dependent induction of drug-metabolizing enzymes.

Large-scale evaluation of cytochrome P450 2C9 mediated drug interaction potential with machine learning-based consensus modeling

Cytochrome P450 (CYP) enzymes play an important role in the metabolism of xenobiotics. Since they are connected to drug interactions, screening for potential inhibitors is of utmost importance in drug discovery settings. Our study provides an extensive classification model for P450-drug interactions with one of the most prominent members, the 2C9 isoenzyme. Our model involved the largest set of 45,000 molecules ever used for developing prediction models. The models are based on three different types of descriptors, (a) typical one, two and three dimensional molecular descriptors, (b) chemical and pharmacophore fingerprints and (c) interaction fingerprints with docking scores. Two machine learning algorithms, the boosted tree and the multilayer feedforward of resilient backpropagation network were used and compared based on their performances. The models were validated both internally and using external validation sets. The results showed that the consensus voting technique with custom probability thresholds could provide promising results even in large-scale cases without any restrictions on the applicability domain. Our best model was capable to predict the 2C9 inhibitory activity with the area under the receiver operating characteristic curve (AUC) of 0.85 and 0.84 for the internal and the external test sets, respectively. The chemical space covered with the largest available dataset has reached its limit encompassing publicly available bioactivity data for the 2C9 isoenzyme.

Computational prediction of cytochrome P450 inhibition and induction

Cytochrome P450 (CYP) enzymes play an important role in the phase I metabolism of many xenobiotics. Most drug-drug interactions (DDIs) associated with CYP are caused by either CYP inhibition or induction. The early detection of potential DDIs is highly desirable in the pharmaceutical industry because DDIs can cause serious adverse events, which can lead to poor patient health and drug development failures. Recently, many computational studies predicting CYP inhibition and induction have been reported. The current computational modeling approaches for CYP metabolism are classified as ligand- and structure-based; various techniques, such as quantitative structure-activity relationships, machine learning, docking, and molecular dynamic simulation, are involved in both the approaches. Recently, combining these two approaches have resulted in improvements in the prediction accuracy of DDIs. In this review, we present important, recent developments in the computational prediction of the inhibition of four clinically crucial CYP isoforms (CYP1A2, 2C9, 2D6, and 3A4) and three nuclear receptors (aryl hydrocarbon receptor, constitutive androstane receptor, and pregnane X receptor) involved in the induction of CYP1A2, 2B6, and 3A4, respectively.

Design and Synthesis of Novel 1,3-Thiazole and 2-Hydrazinyl-1,3-Thiazole Derivatives as Anti-Candida Agents: In Vitro Antifungal Screening, Molecular Docking Study, and Spectroscopic Investigation of their Binding Interaction with Bovine Serum Albumin

In the context of there being a limited number of clinically approved drugs for the treatment of Candida sp.-based infections, along with the rapid development of resistance to the existing antifungals, two novel series of 4-phenyl-1,3-thiazole and 2-hydrazinyl-4-phenyl-1,3-thiazole derivatives were synthesized and tested in vitro for their anti-Candida potential. Two compounds (7a and 7e) showed promising inhibitory activity against the pathogenic C. albicans strain, exhibiting substantially lower MIC values (7.81 μg/mL and 3.9 μg/mL, respectively) as compared with the reference drug fluconazole (15.62 μg/mL). Their anti-Candida activity is also supported by molecular docking studies, using the fungal lanosterol C14α-demethylase as the target enzyme. The interaction of the most biologically active synthesized compound 7e with bovine serum albumin was investigated through fluorescence spectroscopy, and the obtained data suggested that this molecule might efficiently bind carrier proteins in vivo in order to reach the target site.

Prediction of Drug-Drug Interactions Related to Inhibition or Induction of Drug-Metabolizing Enzymes

Drug-drug interaction (DDI) is the phenomenon of alteration of the pharmacological activity of a drug(s) when another drug(s) is co-administered in cases of so-called polypharmacy. There are three types of DDIs: pharmacokinetic (PK), pharmacodynamic, and pharmaceutical. PK is the most frequent type of DDI, which often appears as a result of the inhibition or induction of drug-metabolising enzymes (DME). In this review, we summarise in silico methods that may be applied for the prediction of the inhibition or induction of DMEs and describe appropriate computational methods for DDI prediction, showing the current situation and perspectives of these approaches in medicinal and pharmaceutical chemistry. We review sources of information on DDI, which can be used in pharmaceutical investigations and medicinal practice and/or for the creation of computational models. The problem of the inaccuracy and redundancy of these data are discussed. We provide information on the state-of-the-art physiologically- based pharmacokinetic modelling (PBPK) approaches and DME-based in silico methods. In the section on ligand-based methods, we describe pharmacophore models, molecular field analysis, quantitative structure-activity relationships (QSAR), and similarity analysis applied to the prediction of DDI related to the inhibition or induction of DME. In conclusion, we discuss the problems of DDI severity assessment, mention factors that influence severity, and highlight the issues, perspectives and practical using of in silico methods.

Exploring the Chemical Space of Cytochrome P450 Inhibitors Using Integrated Physicochemical Parameters, Drug Efficiency Metrics and Decision Tree Models

The cytochrome P450s (CYPs) play a central role in the metabolism of various endogenous and exogenous compounds including drugs. CYPs are vulnerable to inhibition and induction which can lead to adverse drug reactions. Therefore, insights into the underlying mechanism of CYP450 inhibition and the estimation of overall CYP inhibitor properties might serve as valuable tools during the early phases of drug discovery. Herein, we present a large data set of inhibitors against five major metabolic CYPs (CYP1A2, CYP2C9, CYP2C19, CYP2D6 and CYP3A4) for the evaluation of important physicochemical properties and ligand efficiency metrics to define property trends across various activity levels (active, efficient and inactive). Decision tree models for CYP inhibition were developed with an accuracy >90% for both the training set and 10-folds cross validation. Overall, molecular weight (MW), hydrogen bond acceptors/donors (HBA/HBD) and lipophilicity (clogP/logPo/w) represent important physicochemical descriptors for CYP450 inhibitors. However, highly efficient CYP inhibitors show mean MW, HBA, HBD and logP values between 294.18–482.40,5.0–8.2,1–7.29 and 1.68–2.57, respectively. Our results might help in optimization of toxicological profiles associated with new chemical entities (NCEs), through a better understanding of inhibitor properties leading to CYP-mediated interactions.

Survey of Machine Learning Techniques for Prediction of the Isoform Specificity of Cytochrome P450 Substrates

Background:

Determination or prediction of the Absorption, Distribution, Metabolism, and Excretion (ADME) properties of drug candidates and drug-induced toxicity plays crucial roles in drug discovery and development. Metabolism is one of the most complicated pharmacokinetic properties to be understood and predicted. However, experimental determination of the substrate binding, selectivity, sites and rates of metabolism is time- and recourse- consuming. In the phase I metabolism of foreign compounds (i.e., most of drugs), cytochrome P450 enzymes play a key role. To help develop drugs with proper ADME properties, computational models are highly desired to predict the ADME properties of drug candidates, particularly for drugs binding to cytochrome P450.

Objective:

This narrative review aims to briefly summarize machine learning techniques used in the prediction of the cytochrome P450 isoform specificity of drug candidates.

Results:

Both single-label and multi-label classification methods have demonstrated good performance on modelling and prediction of the isoform specificity of substrates based on their quantitative descriptors.

Conclusion:

This review provides a guide for researchers to develop machine learning-based methods to predict the cytochrome P450 isoform specificity of drug candidates.

Predicting drug-target interactions using Lasso with random forest based on evolutionary information and chemical structure

The identification of drug-target interactions has great significance for pharmaceutical scientific research. Since traditional experimental methods identifying drug-target interactions is costly and time-consuming, the use of machine learning methods to predict potential drug-target interactions has attracted widespread attention. This paper presents a novel drug-target interactions prediction method called LRF-DTIs. Firstly, the pseudo-position specific scoring matrix (PsePSSM) and FP2 molecular fingerprinting were used to extract the features of drug-target. Secondly, using Lasso to reduce the dimension of the extracted feature information and then the Synthetic Minority Oversampling Technique (SMOTE) method was used to deal with unbalanced data. Finally, the processed feature vectors were input into a random forest (RF) classifier to predict drug-target interactions. Through 10 trials of 5-fold cross-validation, the overall prediction accuracies on the enzyme, ion channel (IC), G-protein-coupled receptor (GPCR) and nuclear receptor (NR) datasets reached 98.09%, 97.32%, 95.69%, and 94.88%, respectively, and compared with other prediction methods. In addition, we have tested and verified that our method not only could be applied to predict the new interactions but also could obtain a satisfactory result on the new dataset. All the experimental results indicate that our method can significantly improve the prediction accuracy of drug-target interactions and play a vital role in the new drug research and target protein development. The source code and all datasets are available at https://github.com/QUST-AIBBDRC/LRF-DTIs/ for academic use.

The current limits in virtual screening and property prediction

Beyond finding inhibitors that show high binding affinity to the respective target, there is the challenge of optimizing their properties with respect to metabolic and toxicological issues, as well as further off-target effects. To reduce the experimental effort of synthesizing and testing actual substances in corresponding assays, virtual screening has become an indispensable toolbox in preclinical development. The scope of application covers the prediction of molecular properties including solubility, metabolic liability and binding to antitargets, such as the hERG channel. Furthermore, prediction of binding sites and drugable targets are emerging aspects of virtual screening. Issues involved with the currently applied computational models including machine learning algorithms are outlined, such as limitations to the accuracy of prediction and overfitting.

Screening of cytochrome P450 3A4 inhibitors via in silico and in vitro approaches

Cytochrome P450 3A4 (CYP3A4) is an important member of the CYP family and responsible for metabolizing a broad range of drugs. It is necessary to establish virtual screening models for predicting CYP3A4 inhibitors.

Simulation of the Bottleneck Controlling Access into a Rieske Active Site: Predicting Substrates of Naphthalene 1,2-Dioxygenase

Naphthalene 1,2-dioxygenase (NDO) has been computationally understudied despite the extensive experimental knowledge obtained for this enzyme, including numerous crystal structures and over 100 demonstrated substrates. In this study, we have developed a substrate prediction model that moves away from the traditional active-site-centric approach to include the energetics of substrate entry into the active site. By comparison with experimental data, the accuracy of the model for predicting substrate oxidation is 92%, with a positive predictive value of 93% and a negative predictive value of 98%. Also, the present analysis has revealed that the amino acid residues that provided the largest energetic barrier for compounds entering the active site are residues F224, L227, P234, and L235. In addition, F224 is proposed to play a role in controlling ligand entrance via π-π stacking stabilization as well as providing stabilization via T-shaped π-π interactions once the ligand has reached the active-site cavity. Overall, we present a method capable of being scaled to computationally discover thousands of substrates of NDO, and we present parameters to be used for expanding the prediction method to other members of the Rieske non-heme iron oxygenase family.

In Silico Prediction of Cytochrome P450-Drug Interaction: QSARs for CYP3A4 and CYP2C9

Cytochromes P450 (CYP) are the main actors in the oxidation of xenobiotics and play a crucial role in drug safety, persistence, bioactivation, and drug-drug/food-drug interaction. This work aims to develop Quantitative Structure-Activity Relationship (QSAR) models to predict the drug interaction with two of the most important CYP isoforms, namely 2C9 and 3A4. The presented models are calibrated on 9122 drug-like compounds, using three different modelling approaches and two types of molecular description (classical molecular descriptors and binary fingerprints). For each isoform, three classification models are presented, based on a different approach and with different advantages: (1) a very simple and interpretable classification tree; (2) a local (k-Nearest Neighbor) model based classical descriptors and; (3) a model based on a recently proposed local classifier (N-Nearest Neighbor) on binary fingerprints. The salient features of the work are (1) the thorough model validation and the applicability domain assessment; (2) the descriptor interpretation, which highlighted the crucial aspects of P450-drug interaction; and (3) the consensus aggregation of models, which largely increased the prediction accuracy.

Towards cheminformatics-based estimation of drug therapeutic index: Predicting the protective index of anticonvulsants using a new quantitative structure-index relationship approach

The overall efficacy and safety profile of a new drug is partially evaluated by the therapeutic index in clinical studies and by the protective index (PI) in preclinical studies. In-silico predictive methods may facilitate the assessment of these indicators. Although QSAR and QSTR models can be used for predicting PI, their predictive capability has not been evaluated. To test this capability, we developed QSAR and QSTR models for predicting the activity and toxicity of anticonvulsants at accuracy levels above the literature-reported threshold (LT) of good QSAR models as tested by both the internal 5-fold cross validation and external validation method. These models showed significantly compromised PI predictive capability due to the cumulative errors of the QSAR and QSTR models. Therefore, in this investigation a new quantitative structure-index relationship (QSIR) model was devised and it showed improved PI predictive capability that superseded the LT of good QSAR models. The QSAR, QSTR and QSIR models were developed using support vector regression (SVR) method with the parameters optimized by using the greedy search method. The molecular descriptors relevant to the prediction of anticonvulsant activities, toxicities and PIs were analyzed by a recursive feature elimination method. The selected molecular descriptors are primarily associated with the drug-like, pharmacological and toxicological features and those used in the published anticonvulsant QSAR and QSTR models. This study suggested that QSIR is useful for estimating the therapeutic index of drug candidates.

Characterization of differences in substrate specificity among CYP1A1, CYP1A2 and CYP1B1: an integrated approach employing molecular docking and molecular dynamics simulations

Recent trends in new drug discovery of anticancer drugs have made oncologists more aware of the fact that the new drug discovery must target the developing mechanism of tumorigenesis to improve the therapeutic efficacy of antineoplastic drugs. The drugs designed are expected to have high affinity towards the novel targets selectively. Current research highlights overexpression of CYP450s, particularly cytochrome P450 1A1 (CYP1A1), in tumour cells, representing a novel target for anticancer therapy. However, the CYP1 family is identified as posing significant problems in selectivity of anticancer molecules towards CYP1A1. Three members have been identified in the human CYP1 family: CYP1A1, CYP1A2 and CYP1B1. Although sequences of the three isoform have high sequence identity, they have distinct substrate specificities. The understanding of macromolecular features that govern substrate specificity is required to understand the interplay between the protein function and dynamics, design novel antitumour compounds that could be specifically metabolized by only CYP1A1 to mediate their antitumour activity and elucidate the reasons for differences in substrate specificity profile among the three proteins. In the present study, we employed a combination of computational methodologies: molecular docking and molecular dynamics simulations. We utilized eight substrates for elucidating the difference in substrate specificity of the three isoforms. Lastly, we conclude that the substrate specificity of a particular substrate depends upon the type of the active site residues, the dynamic motions in the protein structure upon ligand binding and the physico-chemical characteristics of a particular ligand. Copyright © 2016 John Wiley & Sons, Ltd.

Recent progresses in the exploration of machine learning methods as in-silico ADME prediction tools

In-silico methods have been explored as potential tools for assessing ADME and ADME regulatory properties particularly in early drug discovery stages. Machine learning methods, with their ability in classifying diverse structures and complex mechanisms, are well suited for predicting ADME and ADME regulatory properties. Recent efforts have been directed at the broadening of application scopes and the improvement of predictive performance with particular focuses on the coverage of ADME properties, and exploration of more diversified training data, appropriate molecular features, and consensus modeling. Moreover, several online machine learning ADME prediction servers have emerged. Here we review these progresses and discuss the performances, application prospects and challenges of exploring machine learning methods as useful tools in predicting ADME and ADME regulatory properties.

Improving Detection of Arrhythmia Drug-Drug Interactions in Pharmacovigilance Data through the Implementation of Similarity-Based Modeling

Identification of Drug-Drug Interactions (DDIs) is a significant challenge during drug development and clinical practice. DDIs are responsible for many adverse drug effects (ADEs), decreasing patient quality of life and causing higher care expenses. DDIs are not systematically evaluated in pre-clinical or clinical trials and so the FDA U. S. Food and Drug Administration relies on post-marketing surveillance to monitor patient safety. However, existing pharmacovigilance algorithms show poor performance for detecting DDIs exhibiting prohibitively high false positive rates. Alternatively, methods based on chemical structure and pharmacological similarity have shown promise in adverse drug event detection. We hypothesize that the use of chemical biology data in a post hoc analysis of pharmacovigilance results will significantly improve the detection of dangerous interactions. Our model integrates a reference standard of DDIs known to cause arrhythmias with drug similarity data. To compare similarity between drugs we used chemical structure (both 2D and 3D molecular structure), adverse drug side effects, chemogenomic targets, drug indication classes, and known drug-drug interactions. We evaluated the method on external reference standards. Our results showed an enhancement of sensitivity, specificity and precision in different top positions with the use of similarity measures to rank the candidates extracted from pharmacovigilance data. For the top 100 DDI candidates, similarity-based modeling yielded close to twofold precision enhancement compared to the proportional reporting ratio (PRR). Moreover, the method helps in the DDI decision making through the identification of the DDI in the reference standard that generated the candidate.

In Silico Estimation of Chemical Carcinogenicity with Binary and Ternary Classification Methods

Carcinogenicity is one of the most concerned properties of chemicals to human health, thus it is important to identify chemical carcinogenicity as early as possible. In this study, 829 diverse compounds with rat carcinogenicity were collected from Carcinogenic Potency Database (CPDB). Using six types of fingerprints to represent the molecules, 30 binary and ternary classification models were generated to predict chemical carcinogenicity by five machine learning methods. The models were evaluated by an external validation set containing 87 chemicals from ISSCAN database. The best binary model was developed by MACCS keys and kNN algorithm with predictive accuracy at 83.91 %, while the best ternary model was also generated by MACCS keys and kNN algorithm with overall accuracy at 80.46 %. Furthermore, the best binary and ternary classification models were used to estimate carcinogenicity of tobacco smoke components containing 2251 compounds. 981 ones were predicted as carcinogens by binary classification model, while 110 compounds were predicted as strong carcinogens and 807 ones as weak carcinogens by ternary classification model. The results indicated that our models would be helpful for prediction of chemical carcinogenicity.

A lazy learning-based QSAR classification study for screening potential histone deacetylase 8 (HDAC8) inhibitors

Histone deacetylases 8 (HDAC8) is an enzyme repressing the transcription of various genes including tumour suppressor gene and has already become a target of human cancer treatment. In an effort to facilitate the discovery of HDAC8 inhibitors, two quantitative structure-activity relationship (QSAR) classification models were developed using K nearest neighbours (KNN) and neighbourhood classifier (NEC). Molecular descriptors were calculated for the data set and database compounds using ADRIANA.Code of Molecular Networks. Principal components analysis (PCA) was used to select the descriptors. The developed models were validated by leave-one-out cross validation (LOO CV). The performances of the developed models were evaluated with an external test set. Highly predictive models were used for database virtual screening. Furthermore, hit compounds were subsequently subject to molecular docking. Five hits were obtained based on consensus scoring function and binding affinity as potential HDAC8 inhibitors. Finally, HDAC8 structures in complex with five hits were also subjected to 5 ns molecular dynamics (MD) simulations to evaluate the complex structure stability. To the best of our knowledge, the NEC classification model used in this study is the first application of NEC to virtual screening for drug discovery.

In silico prediction of chemical acute oral toxicity using multi-classification methods

Chemical acute oral toxicity is an important end point in drug design and environmental risk assessment. However, it is difficult to determine by experiments, and in silico methods are hence developed as an alternative. In this study, a comprehensive data set containing 12, 204 diverse compounds with median lethal dose (LD₅₀) was compiled. These chemicals were classified into four categories, namely categories I, II, III and IV, based on the criterion of the U.S. Environmental Protection Agency (EPA). Then several multiclassification models were developed using five machine learning methods, including support vector machine (SVM), C4.5 decision tree (C4.5), random forest (RF), κ-nearest neighbor (kNN), and naïve Bayes (NB) algorithms, along with MACCS and FP4 fingerprints. One-against-one (OAO) and binary tree (BT) strategies were employed for SVM multiclassification. Performances were measured by two external validation sets containing 1678 and 375 chemicals, separately. The overall accuracy of the MACCS-SVM(OAO) model was 83.0% and 89.9% for external validation sets I and II, respectively, which showed reliable predictive accuracy for each class. In addition, some representative substructures responsible for acute oral toxicity were identified using information gain and substructure frequency analysis methods, which might be very helpful for further study to avoid the toxicity.

Modelling cytochromes P450 binding modes to predict P450 inhibition, metabolic stability and isoform selectivity

The cytochromes P450 (P450) superfamily is a diverse group of enzymes involved in the metabolism of xenobiotics, whose orientations within the catalytic site can lead to different binding modes, namely productive, nonproductive, and inhibitory. This article collects the most recent approaches that individually study P450- ligand interactions, including a novel in silico technology, developed in the framework of the Human Cytochrome P450 Consortium initiative, that provides reliable in silico predictions of P450 inhibition, metabolic stability and isoform selectivity.

Discovery of antibiotics-derived polymers for gene delivery using combinatorial synthesis and cheminformatics modeling

We describe the combinatorial synthesis and cheminformatics modeling of aminoglycoside antibiotics-derived polymers for transgene delivery and expression. Fifty-six polymers were synthesized by polymerizing aminoglycosides with diglycidyl ether cross-linkers. Parallel screening resulted in identification of several lead polymers that resulted in high transgene expression levels in cells. The role of polymer physicochemical properties in determining efficacy of transgene expression was investigated using Quantitative Structure-Activity Relationship (QSAR) cheminformatics models based on Support Vector Regression (SVR) and 'building block' polymer structures. The QSAR model exhibited high predictive ability, and investigation of descriptors in the model, using molecular visualization and correlation plots, indicated that physicochemical attributes related to both, aminoglycosides and diglycidyl ethers facilitated transgene expression. This work synergistically combines combinatorial synthesis and parallel screening with cheminformatics-based QSAR models for discovery and physicochemical elucidation of effective antibiotics-derived polymers for transgene delivery in medicine and biotechnology.

Development and experimental test of support vector machines virtual screening method for searching Src inhibitors from large compound libraries

BACKGROUND

Src plays various roles in tumour progression, invasion, metastasis, angiogenesis and survival. It is one of the multiple targets of multi-target kinase inhibitors in clinical uses and trials for the treatment of leukemia and other cancers. These successes and appearances of drug resistance in some patients have raised significant interest and efforts in discovering new Src inhibitors. Various in-silico methods have been used in some of these efforts. It is desirable to explore additional in-silico methods, particularly those capable of searching large compound libraries at high yields and reduced false-hit rates.

RESULTS

We evaluated support vector machines (SVM) as virtual screening tools for searching Src inhibitors from large compound libraries. SVM trained and tested by 1,703 inhibitors and 63,318 putative non-inhibitors correctly identified 93.53%~ 95.01% inhibitors and 99.81%~ 99.90% non-inhibitors in 5-fold cross validation studies. SVM trained by 1,703 inhibitors reported before 2011 and 63,318 putative non-inhibitors correctly identified 70.45% of the 44 inhibitors reported since 2011, and predicted as inhibitors 44,843 (0.33%) of 13.56M PubChem, 1,496 (0.89%) of 168 K MDDR, and 719 (7.73%) of 9,305 MDDR compounds similar to the known inhibitors.

CONCLUSIONS

SVM showed comparable yield and reduced false hit rates in searching large compound libraries compared to the similarity-based and other machine-learning VS methods developed from the same set of training compounds and molecular descriptors. We tested three virtual hits of the same novel scaffold from in-house chemical libraries not reported as Src inhibitor, one of which showed moderate activity. SVM may be potentially explored for searching Src inhibitors from large compound libraries at low false-hit rates.

Structure based model for the prediction of phospholipidosis induction potential of small molecules

Drug-induced phospholipidosis (PLD), characterized by an intracellular accumulation of phospholipids and formation of concentric lamellar bodies, has raised concerns in the drug discovery community, due to its potential adverse effects. To evaluate the PLD induction potential, 4,161 nonredundant drug-like molecules from the National Institutes of Health Chemical Genomics Center (NCGC) Pharmaceutical Collection (NPC), the Library of Pharmacologically Active Compounds (LOPAC), and the Tocris Biosciences collection were screened in a quantitative high-throughput screening (qHTS) format. The potential of drug-lipid complex formation can be linked directly to the structures of drug molecules, and many PLD inducing drugs were found to share common structural features. Support vector machine (SVM) models were constructed by using customized atom types or Molecular Operating Environment (MOE) 2D descriptors as structural descriptors. Either the compounds from LOPAC or randomly selected from the entire data set were used as the training set. The impact of training data with biased structural features and the impact of molecule descriptors emphasizing whole-molecule properties or detailed functional groups at the atom level on model performance were analyzed and discussed. Rebalancing strategies were applied to improve the predictive power of the SVM models. Using the undersampling method, the consensus model using one-third of the compounds randomly selected from the data set as the training set achieved high accuracy of 0.90 in predicting the remaining two-thirds of the compounds constituting the test set, as measured by the area under the receiver operator characteristic curve (AUC-ROC).

RS-Predictor models augmented with SMARTCyp reactivities: robust metabolic regioselectivity predictions for nine CYP isozymes

RS-Predictor is a tool for creating pathway-independent, isozyme-specific, site of metabolism (SOM) prediction models using any set of known cytochrome P450 (CYP) substrates and metabolites. Until now, the RS-Predictor method was only trained and validated on CYP 3A4 data, but in the present study, we report on the versatility the RS-Predictor modeling paradigm by creating and testing regioselectivity models for substrates of the nine most important CYP isozymes. Through curation of source literature, we have assembled 680 substrates distributed among CYPs 1A2, 2A6, 2B6, 2C19, 2C8, 2C9, 2D6, 2E1, and 3A4, the largest publicly accessible collection of P450 ligands and metabolites released to date. A comprehensive investigation into the importance of different descriptor classes for identifying the regioselectivity mediated by each isozyme is made through the generation of multiple independent RS-Predictor models for each set of isozyme substrates. Two of these models include a density functional theory (DFT) reactivity descriptor derived from SMARTCyp. Optimal combinations of RS-Predictor and SMARTCyp are shown to have stronger performance than either method alone, while also exceeding the accuracy of the commercial regioselectivity prediction methods distributed by Optibrium and Schrödinger, correctly identifying a large proportion of the metabolites in each substrate set within the top two rank-positions: 1A2 (83.0%), 2A6 (85.7%), 2B6 (82.1%), 2C19 (86.2%), 2C8 (83.8%), 2C9 (84.5%), 2D6 (85.9%), 2E1 (82.8%), 3A4 (82.3%), and merged (86.0%). Comprehensive datamining of each substrate set and careful statistical analyses of the predictions made by the different models revealed new insights into molecular features that control metabolic regioselectivity and enable accurate prospective prediction of likely SOMs.

A support vector machine classification model for benzo[c]phenathridine analogues with toposiomerase-I inhibitory activity

Benzo[c]phenanthridine (BCP) derivatives were identified as topoisomerase I (TOP-I) targeting agents with pronounced antitumor activity. In this study, a support vector machine model was performed on a series of 73 analogues to classify BCP derivatives according to TOP-I inhibitory activity. The best SVM model with total accuracy of 93% for training set was achieved using a set of 7 descriptors identified from a large set via a random forest algorithm. Overall accuracy of up to 87% and a Matthews coefficient correlation (MCC) of 0.71 were obtained after this SVM classifier was validated internally by a test set of 15 compounds. For two external test sets, 89% and 80% BCP compounds, respectively, were correctly predicted. The results indicated that our SVM model could be used as the filter for designing new BCP compounds with higher TOP-I inhibitory activity.

Modeling chemical interaction profiles: II. Molecular docking, spectral data-activity relationship, and structure-activity relationship models for potent and weak inhibitors of cytochrome P450 CYP3A4 isozyme

Polypharmacy increasingly has become a topic of public health concern, particularly as the U.S. population ages. Drug labels often contain insufficient information to enable the clinician to safely use multiple drugs. Because many of the drugs are bio-transformed by cytochrome P450 (CYP) enzymes, inhibition of CYP activity has long been associated with potentially adverse health effects. In an attempt to reduce the uncertainty pertaining to CYP-mediated drug-drug/chemical interactions, an interagency collaborative group developed a consensus approach to prioritizing information concerning CYP inhibition. The consensus involved computational molecular docking, spectral data-activity relationship (SDAR), and structure-activity relationship (SAR) models that addressed the clinical potency of CYP inhibition. The models were built upon chemicals that were categorized as either potent or weak inhibitors of the CYP3A4 isozyme. The categorization was carried out using information from clinical trials because currently available in vitro high-throughput screening data were not fully representative of the in vivo potency of inhibition. During categorization it was found that compounds, which break the Lipinski rule of five by molecular weight, were about twice more likely to be inhibitors of CYP3A4 compared to those, which obey the rule. Similarly, among inhibitors that break the rule, potent inhibitors were 2–3 times more frequent. The molecular docking classification relied on logistic regression, by which the docking scores from different docking algorithms, CYP3A4 three-dimensional structures, and binding sites on them were combined in a unified probabilistic model. The SDAR models employed a multiple linear regression approach applied to binned 1D ¹³C-NMR and 1D ¹⁵N-NMR spectral descriptors. Structure-based and physical-chemical descriptors were used as the basis for developing SAR models by the decision forest method. Thirty-three potent inhibitors and 88 weak inhibitors of CYP3A4 were used to train the models. Using these models, a synthetic majority rules consensus classifier was implemented, while the confidence of estimation was assigned following the percent agreement strategy. The classifier was applied to a testing set of 120 inhibitors not included in the development of the models. Five compounds of the test set, including known strong inhibitors dalfopristin and tioconazole, were classified as probable potent inhibitors of CYP3A4. Other known strong inhibitors, such as lopinavir, oltipraz, quercetin, raloxifene, and troglitazone, were among 18 compounds classified as plausible potent inhibitors of CYP3A4. The consensus estimation of inhibition potency is expected to aid in the nomination of pharmaceuticals, dietary supplements, environmental pollutants, and occupational and other chemicals for in-depth evaluation of the CYP3A4 inhibitory activity. It may serve also as an estimate of chemical interactions via CYP3A4 metabolic pharmacokinetic pathways occurring through polypharmacy and nutritional and environmental exposures to chemical mixtures.

Modeling chemical interaction profiles: I. Spectral data-activity relationship and structure-activity relationship models for inhibitors and non-inhibitors of cytochrome P450 CYP3A4 and CYP2D6 isozymes

An interagency collaboration was established to model chemical interactions that may cause adverse health effects when an exposure to a mixture of chemicals occurs. Many of these chemicals—drugs, pesticides, and environmental pollutant—interact at the level of metabolic biotransformations mediated by cytochrome P450 (CYP) enzymes. In the present work, spectral data-activity relationship (SDAR) and structure-activity relationship (SAR) approaches were used to develop machine-learning classifiers of inhibitors and non-inhibitors of the CYP3A4 and CYP2D6 isozymes. The models were built upon 602 reference pharmaceutical compounds whose interactions have been deduced from clinical data, and 100 additional chemicals that were used to evaluate model performance in an external validation (EV) test. SDAR is an innovative modeling approach that relies on discriminant analysis applied to binned nuclear magnetic resonance (NMR) spectral descriptors. In the present work, both 1D ¹³C and 1D ¹⁵N-NMR spectra were used together in a novel implementation of the SDAR technique. It was found that increasing the binning size of 1D ¹³C-NMR and ¹⁵N-NMR spectra caused an increase in the tenfold cross-validation (CV) performance in terms of both the rate of correct classification and sensitivity. The results of SDAR modeling were verified using SAR. For SAR modeling, a decision forest approach involving from 6 to 17 Mold² descriptors in a tree was used. Average rates of correct classification of SDAR and SAR models in a hundred CV tests were 60% and 61% for CYP3A4, and 62% and 70% for CYP2D6, respectively. The rates of correct classification of SDAR and SAR models in the EV test were 73% and 86% for CYP3A4, and 76% and 90% for CYP2D6, respectively. Thus, both SDAR and SAR methods demonstrated a comparable performance in modeling a large set of structurally diverse data. Based on unique NMR structural descriptors, the new SDAR modeling method complements the existing SAR techniques, providing an independent estimator that can increase confidence in a structure-activity assessment. When modeling was applied to hazardous environmental chemicals, it was found that up to 20% of them may be substrates and up to 10% of them may be inhibitors of the CYP3A4 and CYP2D6 isoforms. The developed models provide a rare opportunity for the environmental health branch of the public health service to extrapolate to hazardous chemicals directly from human clinical data. Therefore, the pharmacological and environmental health branches are both expected to benefit from these reported models.

Computational prediction of metabolism: sites, products, SAR, P450 enzyme dynamics, and mechanisms

Metabolism of xenobiotics remains a central challenge for the discovery and development of drugs, cosmetics, nutritional supplements, and agrochemicals. Metabolic transformations are frequently related to the incidence of toxic effects that may result from the emergence of reactive species, the systemic accumulation of metabolites, or by induction of metabolic pathways. Experimental investigation of the metabolism of small organic molecules is particularly resource demanding; hence, computational methods are of considerable interest to complement experimental approaches. This review provides a broad overview of structure- and ligand-based computational methods for the prediction of xenobiotic metabolism. Current computational approaches to address xenobiotic metabolism are discussed from three major perspectives: (i) prediction of sites of metabolism (SOMs), (ii) elucidation of potential metabolites and their chemical structures, and (iii) prediction of direct and indirect effects of xenobiotics on metabolizing enzymes, where the focus is on the cytochrome P450 (CYP) superfamily of enzymes, the cardinal xenobiotics metabolizing enzymes. For each of these domains, a variety of approaches and their applications are systematically reviewed, including expert systems, data mining approaches, quantitative structure–activity relationships (QSARs), and machine learning-based methods, pharmacophore-based algorithms, shape-focused techniques, molecular interaction fields (MIFs), reactivity-focused techniques, protein–ligand docking, molecular dynamics (MD) simulations, and combinations of methods. Predictive metabolism is a developing area, and there is still enormous potential for improvement. However, it is clear that the combination of rapidly increasing amounts of available ligand- and structure-related experimental data (in particular, quantitative data) with novel and diverse simulation and modeling approaches is accelerating the development of effective tools for prediction of in vivo metabolism, which is reflected by the diverse and comprehensive data sources and methods for metabolism prediction reviewed here. This review attempts to survey the range and scope of computational methods applied to metabolism prediction and also to compare and contrast their applicability and performance.

Identification of cytochrome P450 2D6 and 2C9 substrates and inhibitors by QSAR analysis

This paper presents four new QSAR models for CYP2C9 and CYP2D6 substrate recognition and inhibitor identification based on human clinical data. The models were used to screen a large data set of environmental chemicals for CYP activity, and to analyze the frequency of CYP activity among these compounds. A large fraction of these chemicals were found to be CYP active, and thus potentially capable of affecting human physiology. 20% of the compounds within applicability domain of the models were predicted to be CYP2C9 substrates, and 17% to be inhibitors. The corresponding numbers for CYP2D6 were 9% and 21%. Where the majority of CYP2C9 active compounds were predicted to be both a substrate and an inhibitor at the same time, the CYP2D6 active compounds were primarily predicted to be only inhibitors. It was demonstrated that the models could identify compound classes with a high occurrence of specific CYP activity. An overrepresentation was seen for poly-aromatic hydrocarbons (group of procarcinogens) among CYP2C9 active and mutagenic compounds compared to CYP2C9 inactive and mutagenic compounds. The mutagenicity was predicted with a QSAR model based on Ames in vitro test data.

Transfer learning for cytochrome P450 isozyme selectivity prediction

In the drug discovery process, the metabolic fate of drugs is crucially important to prevent drug-drug interactions. Therefore, P450 isozyme selectivity prediction is an important task for screening drugs of appropriate metabolism profiles. Recently, large-scale activity data of five P450 isozymes (CYP1A2 CYP2C9, CYP3A4, CYP2D6, and CYP2C19) have been obtained using quantitative high-throughput screening with a bioluminescence assay. Although some isozymes share similar selectivities, conventional supervised learning algorithms independently learn a prediction model from each P450 isozyme. They are unable to exploit the other P450 isozyme activity data to improve the predictive performance of each P450 isozyme's selectivity. To address this issue, we apply transfer learning that uses activity data of the other isozymes to learn a prediction model from multiple P450 isozymes. After using the large-scale P450 isozyme selectivity dataset for five P450 isozymes, we evaluate the model's predictive performance. Experimental results show that, overall, our algorithm outperforms conventional supervised learning algorithms such as support vector machine (SVM), Weighted k-nearest neighbor classifier, Bagging, Adaboost, and latent semantic indexing (LSI). Moreover, our results show that the predictive performance of our algorithm is improved by exploiting the multiple P450 isozyme activity data in the learning process. Our algorithm can be an effective tool for P450 selectivity prediction for new chemical entities using multiple P450 isozyme activity data.

Selecting Relevant Descriptors for Classification by Bayesian Estimates: A Comparison with Decision Trees and Support Vector Machines Approaches for Disparate Data Sets

Classification algorithms suffer from the curse of dimensionality, which leads to overfitting, particularly if the problem is over-determined. Therefore it is of particular interest to identify the most relevant descriptors to reduce the complexity. We applied Bayesian estimates to model the probability distribution of descriptors values used for binary classification using n-fold cross-validation. As a measure for the discriminative power of the classifiers, the symmetric form of the Kullback-Leibler divergence of their probability distributions was computed. We found that the most relevant descriptors possess a Gaussian-like distribution of their values, show the largest divergences, and therefore appear most often in the cross-validation scenario. The results were compared to those of the LASSO feature selection method applied to multiple decision trees and support vector machine approaches for data sets of substrates and nonsubstrates of three Cytochrome P450 isoenzymes, which comprise strongly unbalanced compound distributions. In contrast to decision trees and support vector machines, the performance of Bayesian estimates is less affected by unbalanced data sets. This strategy reveals those descriptors that allow a simple linear separation of the classes, whereas the superior accuracy of decision trees and support vector machines can be attributed to nonlinear separation, which are in turn more prone to overfitting.

Insights into molecular basis of cytochrome p450 inhibitory promiscuity of compounds

Cytochrome P450 inhibitory promiscuity of a drug has potential effects on the occurrence of clinical drug-drug interactions. Understanding how a molecular property is related to the P450 inhibitory promiscuity could help to avoid such adverse effects. In this study, an entropy-based index was defined to quantify the P450 inhibitory promiscuity of a compound based on a comprehensive data set, containing more than 11,500 drug-like compounds with inhibition against five major P450 isoforms, 1A2, 2C9, 2C19, 2D6, and 3A4. The results indicated that the P450 inhibitory promiscuity of a compound would have a moderate correlation with molecular aromaticity, a minor correlation with molecular lipophilicity, and no relations with molecular complexity, hydrogen bonding ability, and TopoPSA. We also applied an index to quantify the susceptibilities of different P450 isoforms to inhibition based on the same data set. The results showed that there was a surprising level of P450 inhibitory promiscuity even for substrate specific P450, susceptibility to inhibition follows the rank-order: 1A2 > 2C19 > 3A4 > 2C9 > 2D6. There was essentially no correlation between P450 inhibitory potency and specificity and minor negative trade-offs between P450 inhibitory promiscuity and catalytic promiscuity. In addition, classification models were built to predict the P450 inhibitory promiscuity of new chemicals using support vector machine algorithm with different fingerprints. The area under the receiver operating characteristic curve of the best model was about 0.9, evaluated by 5-fold cross-validation. These findings would be helpful for understanding the mechanism of P450 inhibitory promiscuity and improving the P450 inhibitory selectivity of new chemicals in drug discovery.