Virtual screening of GPCRs: an in silico chemogenomics approach

ERNEST COST action overview on the (patho)physiology of GPCRs and orphan GPCRs in the nervous system

G protein-coupled receptors (GPCRs) are a large family of cell surface receptors that play a critical role in nervous system function by transmitting signals between cells and their environment. They are involved in many, if not all, nervous system processes, and their dysfunction has been linked to various neurological disorders representing important drug targets. This overview emphasises the GPCRs of the nervous system, which are the research focus of the members of ERNEST COST action (CA18133) working group 'Biological roles of signal transduction'. First, the (patho)physiological role of the nervous system GPCRs in the modulation of synapse function is discussed. We then debate the (patho)physiology and pharmacology of opioid, acetylcholine, chemokine, melatonin and adhesion GPCRs in the nervous system. Finally, we address the orphan GPCRs, their implication in the nervous system function and disease, and the challenges that need to be addressed to deorphanize them.

The Histone Deacetylase Family: Structural Features and Application of Combined Computational Methods

Histone deacetylases (HDACs) are crucial in gene transcription, removing acetyl groups from histones. They also influence the deacetylation of non-histone proteins, contributing to the regulation of various biological processes. Thus, HDACs play pivotal roles in various diseases, including cancer, neurodegenerative disorders, and inflammatory conditions, highlighting their potential as therapeutic targets. This paper reviews the structure and function of the four classes of human HDACs. While four HDAC inhibitors are currently available for treating hematological malignancies, numerous others are undergoing clinical trials. However, their non-selective toxicity necessitates ongoing research into safer and more efficient class-selective or isoform-selective inhibitors. Computational methods have aided the discovery of HDAC inhibitors with the desired potency and/or selectivity. These methods include ligand-based approaches, such as scaffold hopping, pharmacophore modeling, three-dimensional quantitative structure-activity relationships, and structure-based virtual screening (molecular docking). Moreover, recent developments in the field of molecular dynamics simulations, combined with Poisson-Boltzmann/molecular mechanics generalized Born surface area techniques, have improved the prediction of ligand binding affinity. In this review, we delve into the ways in which these methods have contributed to designing and identifying HDAC inhibitors.

Orphan G protein-coupled receptors: the ongoing search for a home

G protein-coupled receptors (GPCRs) make up the largest receptor superfamily, accounting for 4% of protein-coding genes. Despite the prevalence of such transmembrane receptors, a significant number remain orphans, lacking identified endogenous ligands. Since their conception, the reverse pharmacology approach has been used to characterize such receptors. However, the multifaceted and nuanced nature of GPCR signaling poses a great challenge to their pharmacological elucidation. Considering their therapeutic relevance, the search for native orphan GPCR ligands continues. Despite limited structural input in terms of 3D crystallized structures, with advances in machine-learning approaches, there has been great progress with respect to accurate ligand prediction. Though such an approach proves valuable given that ligand scarcity is the greatest hurdle to orphan GPCR deorphanization, the future pairings of the remaining orphan GPCRs may not necessarily take a one-size-fits-all approach but should be more comprehensive in accounting for numerous nuanced possibilities to cover the full spectrum of GPCR signaling.

Exploring isofunctional molecules: Design of a benchmark and evaluation of prediction performance

Identification of novel chemotypes with biological activity similar to a known active molecule is an important challenge in drug discovery called 'scaffold hopping'. Small-, medium-, and large-step scaffold hopping efforts may lead to increasing degrees of chemical structure novelty with respect to the parent compound. In the present paper, we focus on the problem of large-step scaffold hopping. We assembled a high quality and well characterized dataset of scaffold hopping examples comprising pairs of active molecules and including a variety of protein targets. This dataset was used to build a benchmark corresponding to the setting of real-life applications: one active molecule is known, and the second active is searched among a set of decoys chosen in a way to avoid statistical bias. This allowed us to evaluate the performance of computational methods for solving large-step scaffold hopping problems. In particular, we assessed how difficult these problems are, particularly for classical 2D and 3D ligand-based methods. We also showed that a machine-learning chemogenomic algorithm outperforms classical methods and we provided some useful hints for future improvements.

Ligand-Based Modeling of CXC Chemokine Receptor 4 and Identification of Inhibitors of Novel Chemotypes as Potential Leads towards New Anti-COVID-19 Treatments

BACKGROUND

Chemokines are involved in several human diseases and in different stages of COVID-19 infection and play critical role in the pathophysiology of the associated acute respiratory disease syndrome, a major complication leading to death among COVID-19 patients. In particular, CXC chemokine receptor 4 (CXCR4) was found to be highly expressed in COVID-19 patients.

METHODS

We herein describe a computational workflow based on combining pharmacophore modeling and QSAR analysis towards the discovery of novel CXCR4 inhibitors. Subsequent virtual screening identified two promising CXCR4 inhibitors from the National Cancer Institute (NCI) list of compounds. The most active hit showed in vitro IC50 value of 24.4 µM.

RESULTS AND CONCLUSION

These results prove the validity of the QSAR model and associated pharmacophore models as means to screen virtual databases towards new CXCR4 inhibitors as leads for the development of new COVID-19 therapies.

Predicting Drug-Target Interactions Based on the Ensemble Models of Multiple Feature Pairs

Backgroud: The prediction of drug–target interactions (DTIs) is of great significance in drug development. It is time-consuming and expensive in traditional experimental methods. Machine learning can reduce the cost of prediction and is limited by the characteristics of imbalanced datasets and problems of essential feature selection. Methods: The prediction method based on the Ensemble model of Multiple Feature Pairs (Ensemble-MFP) is introduced. Firstly, three negative sets are generated according to the Euclidean distance of three feature pairs. Then, the negative samples of the validation set/test set are randomly selected from the union set of the three negative sets in the validation set/test set. At the same time, the ensemble model with weight is optimized and applied to the test set. Results: The area under the receiver operating characteristic curve (area under ROC, AUC) in three out of four sub-datasets in gold standard datasets was more than 94.0% in the prediction of new drugs. The effectiveness of the proposed method is also shown with the comparison of state-of-the-art methods and demonstration of predicted drug–target pairs. Conclusion: The Ensemble-MFP can weigh the existing feature pairs and has a good prediction effect for general prediction on new drugs.

Understanding Membrane Protein Drug Targets in Computational Perspective

Membrane proteins play crucial physiological roles in vivo and are the major category of drug targets for pharmaceuticals. The research on membrane protein is a significant part in the drug discovery. The biological process is a cycled network, and the membrane protein is a vital hub in the network since most drugs achieve the therapeutic effect via interacting with the membrane protein. In this review, typical membrane protein targets are described, including GPCRs, transporters and ion channels. Also, we conclude network servers and databases that are referring to the drug, drug-target information and their relevant data. Furthermore, we chiefly introduce the development and practice of modern medicines, particularly demonstrating a series of state-of-the-art computational models for the prediction of drug-target interaction containing network-based approach and machine-learningbased approach as well as showing current achievements. Finally, we discuss the prospective orientation of drug repurposing and drug discovery as well as propose some improved framework in bioactivity data, created or improved predicted approaches, alternative understanding approaches of drugs bioactivity and their biological processes.

Evaluation of deep and shallow learning methods in chemogenomics for the prediction of drugs specificity

Chemogenomics, also called proteochemometrics, covers a range of computational methods that can be used to predict protein–ligand interactions at large scales in the protein and chemical spaces. They differ from more classical ligand-based methods (also called QSAR) that predict ligands for a given protein receptor. In the context of drug discovery process, chemogenomics allows to tackle the question of predicting off-target proteins for drug candidates, one of the main causes of undesirable side-effects and failure within drugs development processes. The present study compares shallow and deep machine-learning approaches for chemogenomics, and explores data augmentation techniques for deep learning algorithms in chemogenomics. Shallow machine-learning algorithms rely on expert-based chemical and protein descriptors, while recent developments in deep learning algorithms enable to learn abstract numerical representations of molecular graphs and protein sequences, in order to optimise the performance of the prediction task. We first propose a formulation of chemogenomics with deep learning, called the chemogenomic neural network (CN), as a feed-forward neural network taking as input the combination of molecule and protein representations learnt by molecular graph and protein sequence encoders. We show that, on large datasets, the deep learning CN model outperforms state-of-the-art shallow methods, and competes with deep methods with expert-based descriptors. However, on small datasets, shallow methods present better prediction performance than deep learning methods. Then, we evaluate data augmentation techniques, namely multi-view and transfer learning, to improve the prediction performance of the chemogenomic neural network. We conclude that a promising research direction is to integrate heterogeneous sources of data such as auxiliary tasks for which large datasets are available, or independently, multiple molecule and protein attribute views.

Advances in the Understanding of the Cannabinoid Receptor 1 – Focusing on the Inverse Agonists Interactions

Background:

Cannabinoid Receptor 1 (CB1) is a membrane protein prevalent in the central nervous system, whose crystallographic structure has recently been solved. Studies will be needed to investigate CB1 complexes with its ligands and its role in the development of new drugs.

Objective:

Our goal here is to review the studies on CB1, starting with general aspects and focusing on the recent structural studies, with emphasis on the inverse agonists bound structures.

Methods:

We start with a literature review, and then we describe recent studies on CB 1 crystallographic structure and docking simulations. We use this structural information to depict protein-ligand interactions. We also describe the molecular docking method to obtain complex structures of CB 1 with inverse agonists.

Results:

Analysis of the crystallographic structure and docking results revealed the residues responsible for the specificity of the inverse agonists for CB 1. Most of the intermolecular interactions involve hydrophobic residues, with the participation of the residues Phe 170 and Leu 359 in all complex structures investigated in the present study. For the complexes with otenabant and taranabant, we observed intermolecular hydrogen bonds involving residues His 178 (otenabant) and Thr 197 and Ser 383 (taranabant).

Conclusion:

Analysis of the structures involving inverse agonists and CB 1 revealed the pivotal role played by residues Phe 170 and Leu 359 in their interactions and the strong intermolecular hydrogen bonds highlighting the importance of the exploration of intermolecular interactions in the development of novel inverse agonists.

Multi-task generative topographic mapping in virtual screening

The previously reported procedure to generate "universal" Generative Topographic Maps (GTMs) of the drug-like chemical space is in practice a multi-task learning process, in which both operational GTM parameters (example: map grid size) and hyperparameters (key example: the molecular descriptor space to be used) are being chosen by an evolutionary process in order to fit/select "universal" GTM manifolds. After selection (a one-time task aimed at optimizing the compromise in terms of neighborhood behavior compliance, over a large pool of various biological targets), for any further use the manifolds are ready to provide "fit-free" predictive models. Using any structure-activity set-irrespectively whether the associated target served at map fitting stage or not-the generation or "coloring" a property landscape enables predicting the property for any external molecule, with zero additional fitable parameters involved. While previous works have signaled the excellent behavior of such models in aggressive three-fold cross-validation assessments of their predictive power, the present work wished to explore their behavior in Virtual Screening (VS), here simulated on hand of external DUD ligand and decoy series that are fully disjoint from the ChEMBL-extracted landscape coloring sets. Beyond the rather robust results of the universal GTM manifolds in this challenge, it could be shown that the descriptor spaces selected by the evolutionary multi-task learner were intrinsically able to serve as an excellent support for many other VS procedures, starting from parameter-free similarity searching, to local (target-specific) GTM models, to parameter-rich, nonlinear Random Forest and Neural Network approaches.

Efficient multi-task chemogenomics for drug specificity prediction

Adverse drug reactions, also called side effects, range from mild to fatal clinical events and significantly affect the quality of care. Among other causes, side effects occur when drugs bind to proteins other than their intended target. As experimentally testing drug specificity against the entire proteome is out of reach, we investigate the application of chemogenomics approaches. We formulate the study of drug specificity as a problem of predicting interactions between drugs and proteins at the proteome scale. We build several benchmark datasets, and propose NN-MT, a multi-task Support Vector Machine (SVM) algorithm that is trained on a limited number of data points, in order to solve the computational issues or proteome-wide SVM for chemogenomics. We compare NN-MT to different state-of-the-art methods, and show that its prediction performances are similar or better, at an efficient calculation cost. Compared to its competitors, the proposed method is particularly efficient to predict (protein, ligand) interactions in the difficult double-orphan case, i.e. when no interactions are previously known for the protein nor for the ligand. The NN-MT algorithm appears to be a good default method providing state-of-the-art or better performances, in a wide range of prediction scenario that are considered in the present study: proteome-wide prediction, protein family prediction, test (protein, ligand) pairs dissimilar to pairs in the train set, and orphan cases.

Sparse Modeling to Analyze Drug-Target Interaction Networks

Most drugs produce their phenotypic effects by interacting with target proteins, and understanding the molecular features that underpin drug-target interactions is crucial when designing a novel drug. In this chapter, we introduce the protocols that have driven recent advances in sparse modeling methods for analyzing drug-target interaction networks within a chemogenomic framework. In this approach, the chemical structures of candidate drug compounds are correlated with the genomic sequences of the candidate target proteins. We demonstrate the use of sparse canonical correspondence analysis and sparsity-induced binary classifiers to extract the underlying molecular features that are most strongly involved in drug-target interactions. We focus on drug chemical substructures and protein domains. Workflows for applying these methods are presented, and an application is described in detail. We consider the characteristics of each method and suggest possible directions for future research.

Methods for Virtual Screening of GPCR Targets: Approaches and Challenges

Virtual screening (VS) has become an integral part of the drug discovery process and is a valuable tool for finding novel chemical starting points for GPCR targets. Ligand-based VS makes use of biochemical data for known, active compounds and has been applied successfully to many diverse GPCRs. Recent progress in GPCR X-ray crystallography has made it possible to incorporate detailed structural information into the VS process. This chapter outlines the latest VS techniques along with examples that highlight successful applications of these methods. Best practices for increasing the likelihood of VS success, as well as ongoing challenges, are also discussed.

Beyond the hype: deep neural networks outperform established methods using a ChEMBL bioactivity benchmark set

The increase of publicly available bioactivity data in recent years has fueled and catalyzed research in chemogenomics, data mining, and modeling approaches. As a direct result, over the past few years a multitude of different methods have been reported and evaluated, such as target fishing, nearest neighbor similarity-based methods, and Quantitative Structure Activity Relationship (QSAR)-based protocols. However, such studies are typically conducted on different datasets, using different validation strategies, and different metrics. In this study, different methods were compared using one single standardized dataset obtained from ChEMBL, which is made available to the public, using standardized metrics (BEDROC and Matthews Correlation Coefficient). Specifically, the performance of Naïve Bayes, Random Forests, Support Vector Machines, Logistic Regression, and Deep Neural Networks was assessed using QSAR and proteochemometric (PCM) methods. All methods were validated using both a random split validation and a temporal validation, with the latter being a more realistic benchmark of expected prospective execution. Deep Neural Networks are the top performing classifiers, highlighting the added value of Deep Neural Networks over other more conventional methods. Moreover, the best method ('DNN_PCM') performed significantly better at almost one standard deviation higher than the mean performance. Furthermore, Multi-task and PCM implementations were shown to improve performance over single task Deep Neural Networks. Conversely, target prediction performed almost two standard deviations under the mean performance. Random Forests, Support Vector Machines, and Logistic Regression performed around mean performance. Finally, using an ensemble of DNNs, alongside additional tuning, enhanced the relative performance by another 27% (compared with unoptimized 'DNN_PCM'). Here, a standardized set to test and evaluate different machine learning algorithms in the context of multi-task learning is offered by providing the data and the protocols. Graphical Abstract .

Proteochemometric Modeling of the Interaction Space of Carbonic Anhydrase and its Inhibitors: An Assessment of Structure-based and Sequence-based Descriptors

Due to its physiological and clinical roles, carbonic anhydrase (CA) is one of the most interesting case studies. There are different classes of CAinhibitors including sulfonamides, polyamines, coumarins and dithiocarbamates (DTCs). However, many of them hardly act as a selective inhibitor against a specific isoform. Therefore, finding highly selective inhibitors for different isoforms of CA is still an ongoing project. Proteochemometrics modeling (PCM) is able to model the bioactivity of multiple compounds against different isoforms of a protein. Therefore, it would be extremely applicable when investigating the selectivity of different ligands towards different receptors. Given the facts, we applied PCM to investigate the interaction space and structural properties that lead to the selective inhibition of CA isoforms by some dithiocarbamates. Our models have provided interesting structural information that can be considered to design compounds capable of inhibiting different isoforms of CA in an improved selective manner. Validity and predictivity of the models were confirmed by both internal and external validation methods; while Y-scrambling approach was applied to assess the robustness of the models. To prove the reliability and the applicability of our findings, we showed how ligands-receptors selectivity can be affected by removing any of these critical findings from the modeling process.

Target prediction utilising negative bioactivity data covering large chemical space

BACKGROUND

In silico analyses are increasingly being used to support mode-of-action investigations; however many such approaches do not utilise the large amounts of inactive data held in chemogenomic repositories. The objective of this work is concerned with the integration of such bioactivity data in the target prediction of orphan compounds to produce the probability of activity and inactivity for a range of targets. To this end, a novel human bioactivity data set was constructed through the assimilation of over 195 million bioactivity data points deposited in the ChEMBL and PubChem repositories, and the subsequent application of a sphere-exclusion selection algorithm to oversample presumed inactive compounds.

RESULTS

A Bernoulli Naïve Bayes algorithm was trained using the data and evaluated using fivefold cross-validation, achieving a mean recall and precision of 67.7 and 63.8 % for active compounds and 99.6 and 99.7 % for inactive compounds, respectively. We show the performances of the models are considerably influenced by the underlying intraclass training similarity, the size of a given class of compounds, and the degree of additional oversampling. The method was also validated using compounds extracted from WOMBAT producing average precision-recall AUC and BEDROC scores of 0.56 and 0.85, respectively. Inactive data points used for this test are based on presumed inactivity, producing an approximated indication of the true extrapolative ability of the models. A distance-based applicability domain analysis was also conducted; indicating an average Tanimoto Coefficient distance of 0.3 or greater between a test and training set can be used to give a global measure of confidence in model predictions. A final comparison to a method trained solely on active data from ChEMBL performed with precision-recall AUC and BEDROC scores of 0.45 and 0.76.

CONCLUSIONS

The inclusion of inactive data for model training produces models with superior AUC and improved early recognition capabilities, although the results from internal and external validation of the models show differing performance between the breadth of models. The realised target prediction protocol is available at https://github.com/lhm30/PIDGIN.Graphical abstractThe inclusion of large scale negative training data for in silico target prediction improves the precision and recall AUC and BEDROC scores for target models.

Improved large-scale prediction of growth inhibition patterns using the NCI60 cancer cell line panel

MOTIVATION

Recent large-scale omics initiatives have catalogued the somatic alterations of cancer cell line panels along with their pharmacological response to hundreds of compounds. In this study, we have explored these data to advance computational approaches that enable more effective and targeted use of current and future anticancer therapeutics.

RESULTS

We modelled the 50% growth inhibition bioassay end-point (GI50) of 17,142 compounds screened against 59 cancer cell lines from the NCI60 panel (941,831 data-points, matrix 93.08% complete) by integrating the chemical and biological (cell line) information. We determine that the protein, gene transcript and miRNA abundance provide the highest predictive signal when modelling the GI50 endpoint, which significantly outperformed the DNA copy-number variation or exome sequencing data (Tukey's Honestly Significant Difference, P <0.05). We demonstrate that, within the limits of the data, our approach exhibits the ability to both interpolate and extrapolate compound bioactivities to new cell lines and tissues and, although to a lesser extent, to dissimilar compounds. Moreover, our approach outperforms previous models generated on the GDSC dataset. Finally, we determine that in the cases investigated in more detail, the predicted drug-pathway associations and growth inhibition patterns are mostly consistent with the experimental data, which also suggests the possibility of identifying genomic markers of drug sensitivity for novel compounds on novel cell lines.

CONTACT

terez@pasteur.fr; ab454@ac.cam.uk

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Modelling ligand selectivity of serine proteases using integrative proteochemometric approaches improves model performance and allows the multi-target dependent interpretation of features

Serine proteases, implicated in important physiological functions, have a high intra-family similarity, which leads to unwanted off-target effects of inhibitors with insufficient selectivity. However, the availability of sequence and structure data has now made it possible to develop approaches to design pharmacological agents that can discriminate successfully between their related binding sites. In this study, we have quantified the relationship between 12,625 distinct protease inhibitors and their bioactivity against 67 targets of the serine protease family (20,213 data points) in an integrative manner, using proteochemometric modelling (PCM). The benchmarking of 21 different target descriptors motivated the usage of specific binding pocket amino acid descriptors, which helped in the identification of active site residues and selective compound chemotypes affecting compound affinity and selectivity. PCM models performed better than alternative approaches (models trained using exclusively compound descriptors on all available data, QSAR) employed for comparison with R(2)/RMSE values of 0.64 ± 0.23/0.66 ± 0.20 vs. 0.35 ± 0.27/1.05 ± 0.27 log units, respectively. Moreover, the interpretation of the PCM model singled out various chemical substructures responsible for bioactivity and selectivity towards particular proteases (thrombin, trypsin and coagulation factor 10) in agreement with the literature. For instance, absence of a tertiary sulphonamide was identified to be responsible for decreased selective activity (by on average 0.27 ± 0.65 pChEMBL units) on FA10. Among the binding pocket residues, the amino acids (arginine, leucine and tyrosine) at positions 35, 39, 60, 93, 140 and 207 were observed as key contributing residues for selective affinity on these three targets.

Machine learning assisted design of highly active peptides for drug discovery

The discovery of peptides possessing high biological activity is very challenging due to the enormous diversity for which only a minority have the desired properties. To lower cost and reduce the time to obtain promising peptides, machine learning approaches can greatly assist in the process and even partly replace expensive laboratory experiments by learning a predictor with existing data or with a smaller amount of data generation. Unfortunately, once the model is learned, selecting peptides having the greatest predicted bioactivity often requires a prohibitive amount of computational time. For this combinatorial problem, heuristics and stochastic optimization methods are not guaranteed to find adequate solutions. We focused on recent advances in kernel methods and machine learning to learn a predictive model with proven success. For this type of model, we propose an efficient algorithm based on graph theory, that is guaranteed to find the peptides for which the model predicts maximal bioactivity. We also present a second algorithm capable of sorting the peptides of maximal bioactivity. Extensive analyses demonstrate how these algorithms can be part of an iterative combinatorial chemistry procedure to speed up the discovery and the validation of peptide leads. Moreover, the proposed approach does not require the use of known ligands for the target protein since it can leverage recent multi-target machine learning predictors where ligands for similar targets can serve as initial training data. Finally, we validated the proposed approach in vitro with the discovery of new cationic antimicrobial peptides. Source code freely available at http://graal.ift.ulaval.ca/peptide-design/.

Part of the complexity of drug discovery is the sheer chemical diversity to explore combined to all requirements a compound must meet to become a commercial drug. Hence, it makes sense to automate this chemical exploration endeavor in a wise, informed, and efficient fashion. Here, we focused on peptides as they have properties that make them excellent drug starting points. Machine learning techniques may replace expensive in-vitro laboratory experiments by learning an accurate model of it. However, computational models also suffer from the combinatorial explosion due to the enormous chemical diversity. Indeed, applying the model to every peptides would take an astronomical amount of computer time. Therefore, given a model, is it possible to determine, using reasonable computational time, the peptide that has the best properties and chance for success? This exact question is what motivated our work. We focused on recent advances in kernel methods and machine learning to learn a model that already had excellent results. We demonstrate that this class of model has mathematical properties that makes it possible to rapidly identify and sort the best peptides. Finally, in-vitro and in-silico results are provided to support and validate this theoretical discovery.

Marine natural products as breast cancer resistance protein inhibitors

Breast cancer resistance protein (BCRP) is a protein belonging to the ATP-binding cassette (ABC) transporter superfamily that has clinical relevance due to its multi-drug resistance properties in cancer. BCRP can be associated with clinical cancer drug resistance, in particular acute myelogenous or acute lymphocytic leukemias. The overexpression of BCRP contributes to the resistance of several chemotherapeutic drugs, such as topotecan, methotrexate, mitoxantrone, doxorubicin and daunorubicin. The Food and Drugs Administration has already recognized that BCRP is clinically one of the most important drug transporters, mainly because it leads to a reduction of clinical efficacy of various anticancer drugs through its ATP-dependent drug efflux pump function as well as its apparent participation in drug resistance. This review article aims to summarize the different research findings on marine natural products with BCRP inhibiting activity. In this sense, the potential modulation of physiological targets of BCRP by natural or synthetic compounds offers a great possibility for the discovery of new drugs and valuable research tools to recognize the function of the complex ABC-transporters.

Polypharmacology modelling using proteochemometrics (PCM): recent methodological developments, applications to target families, and future prospects

Proteochemometric (PCM) modelling is a computational method to model the bioactivity of multiple ligands against multiple related protein targets simultaneously.

Machine-learning approaches in drug discovery: methods and applications

During the past decade, virtual screening (VS) has evolved from traditional similarity searching, which utilizes single reference compounds, into an advanced application domain for data mining and machine-learning approaches, which require large and representative training-set compounds to learn robust decision rules. The explosive growth in the amount of public domain-available chemical and biological data has generated huge effort to design, analyze, and apply novel learning methodologies. Here, I focus on machine-learning techniques within the context of ligand-based VS (LBVS). In addition, I analyze several relevant VS studies from recent publications, providing a detailed view of the current state-of-the-art in this field and highlighting not only the problematic issues, but also the successes and opportunities for further advances.

Drug/nondrug classification using Support Vector Machines with various feature selection strategies

In conjunction with the advance in computer technology, virtual screening of small molecules has been started to use in drug discovery. Since there are thousands of compounds in early-phase of drug discovery, a fast classification method, which can distinguish between active and inactive molecules, can be used for screening large compound collections. In this study, we used Support Vector Machines (SVM) for this type of classification task. SVM is a powerful classification tool that is becoming increasingly popular in various machine-learning applications. The data sets consist of 631 compounds for training set and 216 compounds for a separate test set. In data pre-processing step, the Pearson's correlation coefficient used as a filter to eliminate redundant features. After application of the correlation filter, a single SVM has been applied to this reduced data set. Moreover, we have investigated the performance of SVM with different feature selection strategies, including SVM-Recursive Feature Elimination, Wrapper Method and Subset Selection. All feature selection methods generally represent better performance than a single SVM while Subset Selection outperforms other feature selection methods. We have tested SVM as a classification tool in a real-life drug discovery problem and our results revealed that it could be a useful method for classification task in early-phase of drug discovery.

Computational Prediction of DrugTarget Interactions Using Chemical, Biological, and Network Features

Drugtarget interactions (DTIs) are central to current drug discovery processes. Efforts have been devoted to the development of methodology for predicting DTIs and drugtarget interaction networks. Most existing methods mainly focus on the application of information about drug or protein structure features. In the present work, we proposed a computational method for DTI prediction by combining the information from chemical, biological and network properties. The method was developed based on a learning algorithm-random forest (RF) combined with integrated features for predicting DTIs. Four classes of drugtarget interaction networks in humans involving enzymes, ion channels, G-protein-coupled receptors (GPCRs) and nuclear receptors, are independently used for establishing predictive models. The RF models gave prediction accuracy of 93.52 %, 94.84 %, 89.68 % and 84.72 % for four pharmaceutically useful datasets, respectively. The prediction ability of our approach is comparative to or even better than that of other DTI prediction methods. These comparative results demonstrated the relevance of the network topology as source of information for predicting DTIs. Further analysis confirmed that among our top ranked predictions of DTIs, several DTIs are supported by databases, while the others represent novel potential DTIs. We believe that our proposed approach can help to limit the search space of DTIs and provide a new way towards repositioning old drugs and identifying targets.

Ligand efficiency-based support vector regression models for predicting bioactivities of ligands to drug target proteins

The concept of ligand efficiency (LE) indices is widely accepted throughout the drug design community and is frequently used in a retrospective manner in the process of drug development. For example, LE indices are used to investigate LE optimization processes of already-approved drugs and to re-evaluate hit compounds obtained from structure-based virtual screening methods and/or high-throughput experimental assays. However, LE indices could also be applied in a prospective manner to explore drug candidates. Here, we describe the construction of machine learning-based regression models in which LE indices are adopted as an end point and show that LE-based regression models can outperform regression models based on pIC50 values. In addition to pIC50 values traditionally used in machine learning studies based on chemogenomics data, three representative LE indices (ligand lipophilicity efficiency (LLE), binding efficiency index (BEI), and surface efficiency index (SEI)) were adopted, then used to create four types of training data. We constructed regression models by applying a support vector regression (SVR) method to the training data. In cross-validation tests of the SVR models, the LE-based SVR models showed higher correlations between the observed and predicted values than the pIC50-based models. Application tests to new data displayed that, generally, the predictive performance of SVR models follows the order SEI > BEI > LLE > pIC50. Close examination of the distributions of the activity values (pIC50, LLE, BEI, and SEI) in the training and validation data implied that the performance order of the SVR models may be ascribed to the much higher diversity of the LE-based training and validation data. In the application tests, the LE-based SVR models can offer better predictive performance of compound-protein pairs with a wider range of ligand potencies than the pIC50-based models. This finding strongly suggests that LE-based SVR models are better than pIC50-based models at predicting bioactivities of compounds that could exhibit a much higher (or lower) potency.

Computational chemogenomics: is it more than inductive transfer?

High-throughput assays challenge us to extract knowledge from multi-ligand, multi-target activity data. In QSAR, weights are statically fitted to each ligand descriptor with respect to a single endpoint or target. However, computational chemogenomics (CG) has demonstrated benefits of learning from entire grids of data at once, rather than building target-specific QSARs. A possible reason for this is the emergence of inductive knowledge transfer (IT) between targets, providing statistical robustness to the model, with no assumption about the structure of the targets. Relevant protein descriptors in CG should allow one to learn how to dynamically adjust ligand attribute weights with respect to protein structure. Hence, models built through explicit learning (EL) by including protein information, while benefitting from IT enhancement, should provide additional predictive capability, notably for protein deorphanization. This interplay between IT and EL in CG modeling is not sufficiently studied. While IT is likely to occur irrespective of the injected target information, it is not clear whether and when boosting due to EL may occur. EL is only possible if protein description is appropriate to the target set under investigation. The key issue here is the search for evidence of genuine EL exceeding expectations based on pure IT. We explore the problem in the context of Support Vector Regression, using more than 9,400 pKi values of 31 GPCRs, where compound-protein interactions are represented by the concatenation of vectorial descriptions of compounds and proteins. This provides a unified framework to generate both IT-enhanced and potentially EL-enabled models, where the difference is toggled by supplied protein information. For EL-enabled models, protein information includes genuine protein descriptors such as typical sequence-based terms, but also the experimentally determined affinity cross-correlation fingerprints. These latter benchmark the expected behavior of a quasi-ideal descriptor capturing the actual functional protein-protein relatedness, and therefore thought to be the most likely to enable EL. EL- and IT-based methods were benchmarked alongside classical QSAR, with respect to cross-validation and deorphanization challenges. A rational method for projecting benchmarked methodologies into a strategy space is given, in the aims that the projection will provide directions for the types of molecule designs possible using a given methodology. While EL-enabled strategies outperform classical QSARs and favorably compare to similar published results, they are, in all respects evaluated herein, not strongly distinguished from IT-enhanced models. Moreover, EL-enabled strategies failed to prove superior in deorphanization challenges. Therefore, this paper raises caution that, contrary to common belief and intuitive expectation, the benefits of chemogenomics models over classical QSAR are quite possibly due less to the injection of protein-related information, and rather impacted more by the effect of inductive transfer, due to simultaneous learning from all of the modeled endpoints. These results show that the field of protein descriptor research needs further improvements to truly realize the expected benefit of EL.

Benchmarking of protein descriptor sets in proteochemometric modeling (part 2): modeling performance of 13 amino acid descriptor sets

Background

While a large body of work exists on comparing and benchmarking descriptors of molecular structures, a similar comparison of protein descriptor sets is lacking. Hence, in the current work a total of 13 amino acid descriptor sets have been benchmarked with respect to their ability of establishing bioactivity models. The descriptor sets included in the study are Z-scales (3 variants), VHSE, T-scales, ST-scales, MS-WHIM, FASGAI, BLOSUM, a novel protein descriptor set (termed ProtFP (4 variants)), and in addition we created and benchmarked three pairs of descriptor combinations. Prediction performance was evaluated in seven structure-activity benchmarks which comprise Angiotensin Converting Enzyme (ACE) dipeptidic inhibitor data, and three proteochemometric data sets, namely (1) GPCR ligands modeled against a GPCR panel, (2) enzyme inhibitors (NNRTIs) with associated bioactivities against a set of HIV enzyme mutants, and (3) enzyme inhibitors (PIs) with associated bioactivities on a large set of HIV enzyme mutants.

Results

The amino acid descriptor sets compared here show similar performance (<0.1 log units RMSE difference and <0.1 difference in MCC), while errors for individual proteins were in some cases found to be larger than those resulting from descriptor set differences ( > 0.3 log units RMSE difference and >0.7 difference in MCC). Combining different descriptor sets generally leads to better modeling performance than utilizing individual sets. The best performers were Z-scales (3) combined with ProtFP (Feature), or Z-Scales (3) combined with an average Z-Scale value for each target, while ProtFP (PCA8), ST-Scales, and ProtFP (Feature) rank last.

Conclusions

While amino acid descriptor sets capture different aspects of amino acids their ability to be used for bioactivity modeling is still – on average – surprisingly similar. Still, combining sets describing complementary information consistently leads to small but consistent improvement in modeling performance (average MCC 0.01 better, average RMSE 0.01 log units lower). Finally, performance differences exist between the targets compared thereby underlining that choosing an appropriate descriptor set is of fundamental for bioactivity modeling, both from the ligand- as well as the protein side.

Training based on ligand efficiency improves prediction of bioactivities of ligands and drug target proteins in a machine learning approach

Machine learning methods based on ligand-protein interaction data in bioactivity databases are one of the current strategies for efficiently finding novel lead compounds as the first step in the drug discovery process. Although previous machine learning studies have succeeded in predicting novel ligand-protein interactions with high performance, all of the previous studies to date have been heavily dependent on the simple use of raw bioactivity data of ligand potencies measured by IC50, EC50, K(i), and K(d) deposited in databases. ChEMBL provides us with a unique opportunity to investigate whether a machine-learning-based classifier created by reflecting ligand efficiency other than the IC50, EC50, K(i), and Kd values can also offer high predictive performance. Here we report that classifiers created from training data based on ligand efficiency show higher performance than those from data based on IC50 or K(i) values. Utilizing GPCRSARfari and KinaseSARfari databases in ChEMBL, we created IC50- or K(i)-based training data and binding efficiency index (BEI) based training data then constructed classifiers using support vector machines (SVMs). The SVM classifiers from the BEI-based training data showed slightly higher area under curve (AUC), accuracy, sensitivity, and specificity in the cross-validation tests. Application of the classifiers to the validation data demonstrated that the AUCs and specificities of the BEI-based classifiers dramatically increased in comparison with the IC50- or K(i)-based classifiers. The improvement of the predictive power by the BEI-based classifiers can be attributed to (i) the more separated distributions of positives and negatives, (ii) the higher diversity of negatives in the BEI-based training data in a feature space of SVMs, and (iii) a more balanced number of positives and negatives in the BEI-based training data. These results strongly suggest that training data based on ligand efficiency as well as data based on classical IC50, EC50, K(d), and K(i) values are important when creating a classifier using a machine learning approach based on bioactivity data.

Benchmarking of protein descriptor sets in proteochemometric modeling (part 1): comparative study of 13 amino acid descriptor sets

Background

While a large body of work exists on comparing and benchmarking of descriptors of molecular structures, a similar comparison of protein descriptor sets is lacking. Hence, in the current work a total of 13 different protein descriptor sets have been compared with respect to their behavior in perceiving similarities between amino acids. The descriptor sets included in the study are Z-scales (3 variants), VHSE, T-scales, ST-scales, MS-WHIM, FASGAI and BLOSUM, and a novel protein descriptor set termed ProtFP (4 variants). We investigate to which extent descriptor sets show collinear as well as orthogonal behavior via principal component analysis (PCA).

Results

In describing amino acid similarities, MSWHIM, T-scales and ST-scales show related behavior, as do the VHSE, FASGAI, and ProtFP (PCA3) descriptor sets. Conversely, the ProtFP (PCA5), ProtFP (PCA8), Z-Scales (Binned), and BLOSUM descriptor sets show behavior that is distinct from one another as well as both of the clusters above. Generally, the use of more principal components (>3 per amino acid, per descriptor) leads to a significant differences in the way amino acids are described, despite that the later principal components capture less variation per component of the original input data.

Conclusion

In this work a comparison is provided of how similar (and differently) currently available amino acids descriptor sets behave when converting structure to property space. The results obtained enable molecular modelers to select suitable amino acid descriptor sets for structure-activity analyses, e.g. those showing complementary behavior.

Genome-scale screening of drug-target associations relevant to Ki using a chemogenomics approach

The identification of interactions between drugs and target proteins plays a key role in genomic drug discovery. In the present study, the quantitative binding affinities of drug-target pairs are differentiated as a measurement to define whether a drug interacts with a protein or not, and then a chemogenomics framework using an unbiased set of general integrated features and random forest (RF) is employed to construct a predictive model which can accurately classify drug-target pairs. The predictability of the model is further investigated and validated by several independent validation sets. The built model is used to predict drug-target associations, some of which were confirmed by comparing experimental data from public biological resources. A drug-target interaction network with high confidence drug-target pairs was also reconstructed. This network provides further insight for the action of drugs and targets. Finally, a web-based server called PreDPI-Ki was developed to predict drug-target interactions for drug discovery. In addition to providing a high-confidence list of drug-target associations for subsequent experimental investigation guidance, these results also contribute to the understanding of drug-target interactions. We can also see that quantitative information of drug-target associations could greatly promote the development of more accurate models. The PreDPI-Ki server is freely available via: http://sdd.whu.edu.cn/dpiki.

Learning a peptide-protein binding affinity predictor with kernel ridge regression

BACKGROUND

The cellular function of a vast majority of proteins is performed through physical interactions with other biomolecules, which, most of the time, are other proteins. Peptides represent templates of choice for mimicking a secondary structure in order to modulate protein-protein interaction. They are thus an interesting class of therapeutics since they also display strong activity, high selectivity, low toxicity and few drug-drug interactions. Furthermore, predicting peptides that would bind to a specific MHC alleles would be of tremendous benefit to improve vaccine based therapy and possibly generate antibodies with greater affinity. Modern computational methods have the potential to accelerate and lower the cost of drug and vaccine discovery by selecting potential compounds for testing in silico prior to biological validation.

RESULTS

We propose a specialized string kernel for small bio-molecules, peptides and pseudo-sequences of binding interfaces. The kernel incorporates physico-chemical properties of amino acids and elegantly generalizes eight kernels, comprised of the Oligo, the Weighted Degree, the Blended Spectrum, and the Radial Basis Function. We provide a low complexity dynamic programming algorithm for the exact computation of the kernel and a linear time algorithm for it's approximation. Combined with kernel ridge regression and SupCK, a novel binding pocket kernel, the proposed kernel yields biologically relevant and good prediction accuracy on the PepX database. For the first time, a machine learning predictor is capable of predicting the binding affinity of any peptide to any protein with reasonable accuracy. The method was also applied to both single-target and pan-specific Major Histocompatibility Complex class II benchmark datasets and three Quantitative Structure Affinity Model benchmark datasets.

CONCLUSION

On all benchmarks, our method significantly (p-value ≤ 0.057) outperforms the current state-of-the-art methods at predicting peptide-protein binding affinities. The proposed approach is flexible and can be applied to predict any quantitative biological activity. Moreover, generating reliable peptide-protein binding affinities will also improve system biology modelling of interaction pathways. Lastly, the method should be of value to a large segment of the research community with the potential to accelerate the discovery of peptide-based drugs and facilitate vaccine development. The proposed kernel is freely available at http://graal.ift.ulaval.ca/downloads/gs-kernel/.

Chemogenomics of allosteric binding sites in GPCRs

Chemogenomic techniques connect the chemical and biological domains to establish ligand and target relationships not evident from the individual disciplines. Chemogenomics has been applied in lead generation, target classification, focused library design as well as selectivity and polypharmacology profiling. This review describes recent developments structured into ligand-, target- and combined chemogenomic techniques and applications to allosteric GPCR ligands. It also outlines relative strengths and limitations of these techniques and the impact of the increasing crystallographic data.

Chemogenomic approaches to infer drug-target interaction networks

The identification of drug-target interactions from heterogeneous biological data is critical in the drug development. In this chapter, we review recently developed in silico chemogenomic approaches to infer unknown drug-target interactions from chemical information of drugs and genomic information of target proteins. We review several kernel-based statistical methods from two different viewpoints: binary classification and dimension reduction. In the results, we demonstrate the usefulness of the methods on the prediction of drug-target interactions from chemical structure data and genomic sequence data. We also discuss the characteristics of each method, and show some perspectives toward future research direction.

Screening of selective histone deacetylase inhibitors by proteochemometric modeling

BACKGROUND

Histone deacetylase (HDAC) is a novel target for the treatment of cancer and it can be classified into three classes, i.e., classes I, II, and IV. The inhibitors selectively targeting individual HDAC have been proved to be the better candidate antitumor drugs. To screen selective HDAC inhibitors, several proteochemometric (PCM) models based on different combinations of three kinds of protein descriptors, two kinds of ligand descriptors and multiplication cross-terms were constructed in our study.

RESULTS

The results show that structure similarity descriptors are better than sequence similarity descriptors and geometry descriptors in the leftacterization of HDACs. Furthermore, the predictive ability was not improved by introducing the cross-terms in our models. Finally, a best PCM model based on protein structure similarity descriptors and 32-dimensional general descriptors was derived (R2 = 0.9897, Qtest2 = 0.7542), which shows a powerful ability to screen selective HDAC inhibitors.

CONCLUSIONS

Our best model not only predict the activities of inhibitors for each HDAC isoform, but also screen and distinguish class-selective inhibitors and even more isoform-selective inhibitors, thus it provides a potential way to discover or design novel candidate antitumor drugs with reduced side effect.

DrugLogit: logistic discrimination between drugs and nondrugs including disease-specificity by assigning probabilities based on molecular properties

The increasing knowledge of both structure and activity of compounds provides a good basis for enhancing the pharmacological characterization of chemical libraries. In addition, pharmacology can be seen as incorporating both advances from molecular biology as well as chemical sciences, with innovative insight provided from studying target-ligand data from a ligand molecular point of view. Predictions and profiling of libraries of drug candidates have previously focused mainly on certain cases of oral bioavailability. Inclusion of other administration routes and disease-specificity would improve the precision of drug profiling. In this work, recent data are extended, and a probability-based approach is introduced for quantitative and gradual classification of compounds into categories of drugs/nondrugs, as well as for disease- or organ-specificity. Using experimental data of over 1067 compounds and multivariate logistic regressions, the classification shows good performance in training and independent test cases. The regressions have high statistical significance in terms of the robustness of coefficients and 95% confidence intervals provided by a 1000-fold bootstrapping resampling. Besides their good predictive power, the classification functions remain chemically interpretable, containing only one to five variables in total, and the physicochemical terms involved can be easily calculated. The present approach is useful for an improved description and filtering of compound libraries. It can also be applied sequentially or in combinations of filters, as well as adapted to particular use cases. The scores and equations may be able to suggest possible routes for compound or library modification. The data is made available for reuse by others, and the equations are freely accessible at http://hermes.chem.ut.ee/~alfx/druglogit.html.

Prediction of chemical-protein interactions: multitarget-QSAR versus computational chemogenomic methods

Elucidation of chemical-protein interactions (CPI) is the basis of target identification and drug discovery. It is time-consuming and costly to determine CPI experimentally, and computational methods will facilitate the determination of CPI. In this study, two methods, multitarget quantitative structure-activity relationship (mt-QSAR) and computational chemogenomics, were developed for CPI prediction. Two comprehensive data sets were collected from the ChEMBL database for method assessment. One data set consisted of 81 689 CPI pairs among 50 924 compounds and 136 G-protein coupled receptors (GPCRs), while the other one contained 43 965 CPI pairs among 23 376 compounds and 176 kinases. The range of the area under the receiver operating characteristic curve (AUC) for the test sets was 0.95 to 1.0 and 0.82 to 1.0 for 100 GPCR mt-QSAR models and 100 kinase mt-QSAR models, respectively. The AUC of 5-fold cross validation were about 0.92 for both 176 kinases and 136 GPCRs using the chemogenomic method. However, the performance of the chemogenomic method was worse than that of mt-QSAR for the external validation set. Further analysis revealed that there was a high false positive rate for the external validation set when using the chemogenomic method. In addition, we developed a web server named CPI-Predictor, , which is available for free. The methods and tool have potential applications in network pharmacology and drug repositioning.

Computational models for in-vitro anti-tubercular activity of molecules based on high-throughput chemical biology screening datasets

Background

The emergence of Multi-drug resistant tuberculosis in pandemic proportions throughout the world and the paucity of novel therapeutics for tuberculosis have re-iterated the need to accelerate the discovery of novel molecules with anti-tubercular activity. Though high-throughput screens for anti-tubercular activity are available, they are expensive, tedious and time-consuming to be performed on large scales. Thus, there remains an unmet need to prioritize the molecules that are taken up for biological screens to save on cost and time. Computational methods including Machine Learning have been widely employed to build classifiers for high-throughput virtual screens to prioritize molecules for further analysis. The availability of datasets based on high-throughput biological screens or assays in public domain makes computational methods a plausible proposition for building predictive models. In addition, this approach would save significantly on the cost, effort and time required to run high throughput screens.

Results

We show that by using four supervised state-of-the-art classifiers (SMO, Random Forest, Naive Bayes and J48) we are able to generate in-silico predictive models on an extremely imbalanced (minority class ratio: 0.6%) large dataset of anti-tubercular molecules with reasonable AROC (0.6-0.75) and BCR (60-66%) values. Moreover, these models are able to provide 3-4 fold enrichment over random selection.

Conclusions

In the present study, we have used the data from in-vitro screens for anti-tubercular activity from a high-throughput screen available in public domain to build highly accurate classifiers based on molecular descriptors of the molecules. We show that Machine Learning tools can be used to build highly effective predictive models for virtual high-throughput screens to prioritize molecules from large molecular libraries.