Prediction of the Formation of Reactive Metabolites by A Novel Classifier Approach Based on Enrichment Factor Optimization (EFO) as Implemented in the VEGA Program

Improved Detection of Drug-Induced Liver Injury by Integrating Predicted In Vivo and In Vitro Data

Drug-induced liver injury (DILI) has been a significant challenge in drug discovery, often leading to clinical trial failures and necessitating drug withdrawals. Over the last decade, the existing suite of in vitro proxy-DILI assays has generally improved at identifying compounds with hepatotoxicity. However, there is considerable interest in enhancing the in silico prediction of DILI because it allows for evaluating large sets of compounds more quickly and cost-effectively, particularly in the early stages of projects. In this study, we aim to study ML models for DILI prediction that first predict nine proxy-DILI labels and then use them as features in addition to chemical structural features to predict DILI. The features include in vitro (e.g., mitochondrial toxicity, bile salt export pump inhibition) data, in vivo (e.g., preclinical rat hepatotoxicity studies) data, pharmacokinetic parameters of maximum concentration, structural fingerprints, and physicochemical parameters. We trained DILI-prediction models on 888 compounds from the DILI data set (composed of DILIst and DILIrank) and tested them on a held-out external test set of 223 compounds from the DILI data set. The best model, DILIPredictor, attained an AUC-PR of 0.79. This model enabled the detection of the top 25 toxic compounds (2.68 LR+, positive likelihood ratio) compared to models using only structural features (1.65 LR+ score). Using feature interpretation from DILIPredictor, we identified the chemical substructures causing DILI and differentiated cases of DILI caused by compounds in animals but not in humans. For example, DILIPredictor correctly recognized 2-butoxyethanol as nontoxic in humans despite its hepatotoxicity in mice models. Overall, the DILIPredictor model improves the detection of compounds causing DILI with an improved differentiation between animal and human sensitivity and the potential for mechanism evaluation. DILIPredictor required only chemical structures as input for prediction and is publicly available at https://broad.io/DILIPredictor for use via web interface and with all code available for download.

Improved Detection of Drug-Induced Liver Injury by Integrating Predicted in vivo and in vitro Data

Drug-induced liver injury (DILI) has been significant challenge in drug discovery, often leading to clinical trial failures and necessitating drug withdrawals. The existing suite of in vitro proxy-DILI assays is generally effective at identifying compounds with hepatotoxicity. However, there is considerable interest in enhancing in silico prediction of DILI because it allows for the evaluation of large sets of compounds more quickly and cost-effectively, particularly in the early stages of projects. In this study, we aim to study ML models for DILI prediction that first predicts nine proxy-DILI labels and then uses them as features in addition to chemical structural features to predict DILI. The features include in vitro (e.g., mitochondrial toxicity, bile salt export pump inhibition) data, in vivo (e.g., preclinical rat hepatotoxicity studies) data, pharmacokinetic parameters of maximum concentration, structural fingerprints, and physicochemical parameters. We trained DILI-prediction models on 888 compounds from the DILIst dataset and tested on a held-out external test set of 223 compounds from DILIst dataset. The best model, DILIPredictor, attained an AUC-ROC of 0.79. This model enabled the detection of top 25 toxic compounds compared to models using only structural features (2.68 LR+ score). Using feature interpretation from DILIPredictor, we were able to identify the chemical substructures causing DILI as well as differentiate cases DILI is caused by compounds in animals but not in humans. For example, DILIPredictor correctly recognized 2-butoxyethanol as non-toxic in humans despite its hepatotoxicity in mice models. Overall, the DILIPredictor model improves the detection of compounds causing DILI with an improved differentiation between animal and human sensitivity as well as the potential for mechanism evaluation. DILIPredictor is publicly available at https://broad.io/DILIPredictor for use via web interface and with all code available for download and local implementation via https://pypi.org/project/dilipred/.

Extensive Sampling of Molecular Dynamics Simulations to Identify Reliable Protein Structures for Optimized Virtual Screening Studies: The Case of the hTRPM8 Channel

(1) Background: Virtual screening campaigns require target structures in which the pockets are properly arranged for binding. Without these, MD simulations can be used to relax the available target structures, optimizing the fine architecture of their binding sites. Among the generated frames, the best structures can be selected based on available experimental data. Without experimental templates, the MD trajectories can be filtered by energy-based criteria or sampled by systematic analyses. (2) Methods: A blind and methodical analysis was performed on the already reported MD run of the hTRPM8 tetrameric structures; a total of 50 frames underwent docking simulations by using a set of 1000 ligands including 20 known hTRPM8 modulators. Docking runs were performed by LiGen program and involved the frames as they are and after optimization by SCRWL4.0. For each frame, all four monomers were considered. Predictive models were developed by the EFO algorithm based on the sole primary LiGen scores. (3) Results: On average, the MD simulation progressively enhances the performance of the extracted frames, and the optimized structures perform better than the non-optimized frames (EF1% mean: 21.38 vs. 23.29). There is an overall correlation between performances and volumes of the explored pockets and the combination of the best performing frames allows to develop highly performing consensus models (EF1% = 49.83). (4) Conclusions: The systematic sampling of the entire MD run provides performances roughly comparable with those previously reached by using rationally selected frames. The proposed strategy appears to be helpful when the lack of experimental data does not allow an easy selection of the optimal structures for docking simulations. Overall, the reported docking results confirm the relevance of simulating all the monomers of an oligomer structure and emphasize the efficacy of the SCRWL4.0 method to optimize the protein structures for docking calculations.

Artificial Intelligence, Machine Learning, and Deep Learning in Real-Life Drug Design Cases

The discovery and development of drugs is a long and expensive process with a high attrition rate. Computational drug discovery contributes to ligand discovery and optimization, by using models that describe the properties of ligands and their interactions with biological targets. In recent years, artificial intelligence (AI) has made remarkable modeling progress, driven by new algorithms and by the increase in computing power and storage capacities, which allow the processing of large amounts of data in a short time. This review provides the current state of the art of AI methods applied to drug discovery, with a focus on structure- and ligand-based virtual screening, library design and high-throughput analysis, drug repurposing and drug sensitivity, de novo design, chemical reactions and synthetic accessibility, ADMET, and quantum mechanics.

The VEGA suite of programs: an versatile platform for cheminformatics and drug design projects

The purpose of the article is to offer an overview of the latest release of the VEGA suite of programs. This software has been constantly developed and freely released during the last 20 years and has now reached a significant diffusion and technology level as confirmed by the about 22 500 registered users. While being primarily developed for drug design studies, the VEGA package includes cheminformatics and modeling features, which can be fruitfully utilized in various contexts of the computational chemistry. To offer a glimpse of the remarkable potentials of the software, some examples of the implemented features in the cheminformatics field and for structure-based studies are discussed. Finally, the flexible architecture of the VEGA program which can be expanded and customized by plug-in technology or scripting languages will be described focusing attention on the HyperDrive library including highly optimized functions.

AVAILABILITY AND IMPLEMENTATION

The VEGA suite of programs and the source code of the VEGA command-line version are available free of charge for non-profit organizations at http://www.vegazz.net.

MetaTREE, a Novel Database Focused on Metabolic Trees, Predicts an Important Detoxification Mechanism: The Glutathione Conjugation

(1) Background: Data accuracy plays a key role in determining the model performances and the field of metabolism prediction suffers from the lack of truly reliable data. To enhance the accuracy of metabolic data, we recently proposed a manually curated database collected by a meta-analysis of the specialized literature (MetaQSAR). Here we aim to further increase data accuracy by focusing on publications reporting exhaustive metabolic trees. This selection should indeed reduce the number of false negative data. (2) Methods: A new metabolic database (MetaTREE) was thus collected and utilized to extract a dataset for metabolic data concerning glutathione conjugation (MT-dataset). After proper pre-processing, this dataset, along with the corresponding dataset extracted from MetaQSAR (MQ-dataset), was utilized to develop binary classification models using a random forest algorithm. (3) Results: The comparison of the models generated by the two collected datasets reveals the better performances reached by the MT-dataset (MCC raised from 0.63 to 0.67, sensitivity from 0.56 to 0.58). The analysis of the applicability domain also confirms that the model based on the MT-dataset shows a more robust predictive power with a larger applicability domain. (4) Conclusions: These results confirm that focusing on metabolic trees represents a convenient approach to increase data accuracy by reducing the false negative cases. The encouraging performances shown by the models developed by the MT-dataset invites to use of MetaTREE for predictive studies in the field of xenobiotic metabolism.

Combining Different Docking Engines and Consensus Strategies to Design and Validate Optimized Virtual Screening Protocols for the SARS-CoV-2 3CL Protease

The 3CL-Protease appears to be a very promising medicinal target to develop anti-SARS-CoV-2 agents. The availability of resolved structures allows structure-based computational approaches to be carried out even though the lack of known inhibitors prevents a proper validation of the performed simulations. The innovative idea of the study is to exploit known inhibitors of SARS-CoV 3CL-Pro as a training set to perform and validate multiple virtual screening campaigns. Docking simulations using four different programs (Fred, Glide, LiGen, and PLANTS) were performed investigating the role of both multiple binding modes (by binding space) and multiple isomers/states (by developing the corresponding isomeric space). The computed docking scores were used to develop consensus models, which allow an in-depth comparison of the resulting performances. On average, the reached performances revealed the different sensitivity to isomeric differences and multiple binding modes between the four docking engines. In detail, Glide and LiGen are the tools that best benefit from isomeric and binding space, respectively, while Fred is the most insensitive program. The obtained results emphasize the fruitful role of combining various docking tools to optimize the predictive performances. Taken together, the performed simulations allowed the rational development of highly performing virtual screening workflows, which could be further optimized by considering different 3CL-Pro structures and, more importantly, by including true SARS-CoV-2 3CL-Pro inhibitors (as learning set) when available.

Repositioning Dequalinium as Potent Muscarinic Allosteric Ligand by Combining Virtual Screening Campaigns and Experimental Binding Assays

Structure-based virtual screening is a truly productive repurposing approach provided that reliable target structures are available. Recent progresses in the structural resolution of the G-Protein Coupled Receptors (GPCRs) render these targets amenable for structure-based repurposing studies. Hence, the present study describes structure-based virtual screening campaigns with a view to repurposing known drugs as potential allosteric (and/or orthosteric) ligands for the hM₂ muscarinic subtype which was indeed resolved in complex with an allosteric modulator thus allowing a precise identification of this binding cavity. First, a docking protocol was developed and optimized based on binding space concept and enrichment factor optimization algorithm (EFO) consensus approach by using a purposely collected database including known allosteric modulators. The so-developed consensus models were then utilized to virtually screen the DrugBank database. Based on the computational results, six promising molecules were selected and experimentally tested and four of them revealed interesting affinity data; in particular, dequalinium showed a very impressive allosteric modulation for hM₂. Based on these results, a second campaign was focused on bis-cationic derivatives and allowed the identification of other two relevant hM₂ ligands. Overall, the study enhances the understanding of the factors governing the hM₂ allosteric modulation emphasizing the key role of ligand flexibility as well as of arrangement and delocalization of the positively charged moieties.

Combining Molecular Dynamics and Docking Simulations to Develop Targeted Protocols for Performing Optimized Virtual Screening Campaigns on The hTRPM8 Channel

BACKGROUND

There is an increasing interest in TRPM8 ligands of medicinal interest, the rational design of which can be nowadays supported by structure-based in silico studies based on the recently resolved TRPM8 structures. Methods: The study involves the generation of a reliable hTRPM8 homology model, the reliability of which was assessed by a 1.0 μs MD simulation which was also used to generate multiple receptor conformations for the following structure-based virtual screening (VS) campaigns; docking simulations utilized different programs and involved all monomers of the selected frames; the so computed docking scores were combined by consensus approaches based on the EFO algorithm. Results: The obtained models revealed very satisfactory performances; LiGen™ provided the best results among the tested docking programs; the combination of docking results from the four monomers elicited a markedly beneficial effect on the computed consensus models. Conclusions: The generated hTRPM8 model appears to be amenable for successful structure-based VS studies; cross-talk modulating effects between interacting monomers on the binding sites can be accounted for by combining docking simulations as performed on all the monomers; this strategy can have general applicability for docking simulations involving quaternary protein structures with multiple identical binding pockets.

Rescoring and Linearly Combining: A Highly Effective Consensus Strategy for Virtual Screening Campaigns

The study proposes a novel consensus strategy based on linear combinations of different docking scores to be used in the evaluation of virtual screening campaigns. The consensus models are generated by applying the recently proposed Enrichment Factor Optimization (EFO) method, which develops the linear equations by exhaustively combining the available docking scores and by optimizing the resulting enrichment factors. The performances of such a consensus strategy were evaluated by simulating the entire Directory of Useful Decoys (DUD datasets). In detail, the poses were initially generated by the PLANTS docking program and then rescored by ReScore+ with and without the minimization of the complexes. The so calculated scores were then used to generate the mentioned consensus models including two or three different scoring functions. The reliability of the generated models was assessed by a per target validation as performed by default by the EFO approach. The encouraging performances of the here proposed consensus strategy are emphasized by the average increase of the 17% in the Top 1% enrichment factor (EF) values when comparing the single best score with the linear combination of three scores. Specifically, kinases offer a truly convincing demonstration of the efficacy of the here proposed consensus strategy since their Top 1% EF average ranges from 6.4 when using the single best performing primary score to 23.5 when linearly combining scoring functions. The beneficial effects of this consensus approach are clearly noticeable even when considering the entire DUD datasets as evidenced by the area under the curve (AUC) averages revealing a 14% increase when combining three scores. The reached AUC values compare very well with those reported in literature by an extended set of recent benchmarking studies and the three-variable models afford the highest AUC average.