Improving Structure-Based Virtual Screening with Ensemble Docking and Machine Learning

Exploring isoindolin-1-ones as potential CDK7 inhibitors using cheminformatic tools

In women who die from cancer, breast cancer is the most common cause of death. The development of small molecular scaffolds as specific Cyclin-dependent kinase (CDK) inhibitors is a promising strategy in the discovery of anti-breast cancer drugs. Isoindolin-1-ones are heterocyclic compounds with useful therapeutic properties. In this study, a library of 48 isoindolinones has been virtually screened by molecular docking that showed high binding affinity up to - 10.1 kcal/mol and conventional hydrogen bonding interactions with active amino acid residues of CDK7. The molecular dynamics simulation (MDS), fragment molecular orbital (FMO), density functional theory (DFT), and pharmacokinetics studies of the best two docked scored ligands 7 and 14 have been studied. Examining the ligand root mean square deviation and hydrogen bonding occupancy of the 100 ns MDS trajectory, both ligands 7 and 14 showed docked pose stability. FMO calculations displayed that LYS139 and LYS41 are majorly contributing to the binding interactions with ligands 7 and 14 in the docked poses. DFT studies of ligands 7 and 14 showed high values of global softness and low values of global hardness and chemical potential thus displaying chemically reactive soft molecules and this influences their anti-cancer activity. Our hits exhibited superior qualities to known CDK7 inhibitors, according to the comprehensive pharmacokinetic parameters that were predicted. The results indicate that isoindolin-1-one moieties are good candidates for anti-cancer action and could serve as effective CDK7 inhibitors.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40203-024-00225-0.

Predicting binding events in very flexible, allosteric, multi-domain proteins

Knowledge of the structures formed by proteins and small ligands is of fundamental importance for understanding molecular principles of chemotherapy and for designing new and more effective drugs. Due to the still high costs and to the several limitations of experimental techniques, it is most often desirable to predict these ligand-protein complexes in silico, particularly when screening for new putative drugs from databases of millions of compounds. While virtual screening based on molecular docking is widely used for this purpose, it generally fails in mimicking binding events associated with large conformational changes in the protein, particularly when the latter involve multiple domains. In this work, we describe a new methodology aimed at generating bound-like conformations of very flexible and allosteric proteins bearing multiple binding sites. Validation was performed on the enzyme adenylate kinase (ADK), a paradigmatic example of proteins that undergo very large conformational changes upon ligand binding. By only exploiting the unbound structure and the putative binding sites of the protein, we generated a significant fraction of bound-like structures, which employed in ensemble-docking calculations allowed to find native-like poses of substrates, inhibitors, and catalytically incompetent binders. Our protocol provides a general framework for the generation of bound-like conformations of flexible proteins that are suitable to host different ligands, demonstrating high sensitivity to the fine chemical details that regulate protein's activity. We foresee applications in virtual screening for difficult targets, prediction of the impact of amino acid mutations on structure and dynamics, and protein engineering.

Anti-Obesity Therapeutic Targets Studied In Silico and In Vivo: A Systematic Review

In the age of information technology and the additional computational search tools and software available, this systematic review aimed to identify potential therapeutic targets for obesity, evaluated in silico and subsequently validated in vivo. The systematic review was initially guided by the research question "What therapeutic targets have been used in in silico analysis for the treatment of obesity?" and structured based on the acronym PECo (P, problem; E, exposure; Co, context). The systematic review protocol was formulated and registered in PROSPERO (CRD42022353808) in accordance with the Preferred Reporting Items Checklist for Systematic Review and Meta-Analysis Protocols (PRISMA-P), and the PRISMA was followed for the systematic review. The studies were selected according to the eligibility criteria, aligned with PECo, in the following databases: PubMed, ScienceDirect, Scopus, Web of Science, BVS, and EMBASE. The search strategy yielded 1142 articles, from which, based on the evaluation criteria, 12 were included in the systematic review. Only seven these articles allowed the identification of both in silico and in vivo reassessed therapeutic targets. Among these targets, five were exclusively experimental, one was exclusively theoretical, and one of the targets presented an experimental portion and a portion obtained by modeling. The predominant methodology used was molecular docking and the most studied target was Human Pancreatic Lipase (HPL) (n = 4). The lack of methodological details resulted in more than 50% of the papers being categorized with an "unclear risk of bias" across eight out of the eleven evaluated criteria. From the current systematic review, it seems evident that integrating in silico methodologies into studies of potential drug targets for the exploration of new therapeutic agents provides an important tool, given the ongoing challenges in controlling obesity.

Machine learning accelerates pharmacophore-based virtual screening of MAO inhibitors

Nowadays, an efficient and robust virtual screening procedure is crucial in the drug discovery process, especially when performed on large and chemically diverse databases. Virtual screening methods, like molecular docking and classic QSAR models, are limited in their ability to handle vast numbers of compounds and to learn from scarce data, respectively. In this study, we introduce a universal methodology that uses a machine learning-based approach to predict docking scores without the need for time-consuming molecular docking procedures. The developed protocol yielded 1000 times faster binding energy predictions than classical docking-based screening. The proposed predictive model learns from docking results, allowing users to choose their preferred docking software without relying on insufficient and incoherent experimental activity data. The methodology described employs multiple types of molecular fingerprints and descriptors to construct an ensemble model that further reduces prediction errors and is capable of delivering highly precise docking score values for monoamine oxidase ligands, enabling faster identification of promising compounds. An extensive pharmacophore-constrained screening of the ZINC database resulted in a selection of 24 compounds that were synthesized and evaluated for their biological activity. A preliminary screen discovered weak inhibitors of MAO-A with a percentage efficiency index close to a known drug at the lowest tested concentration. The approach presented here can be successfully applied to other biological targets as target-specific knowledge is not incorporated at the screening phase.

PL-PatchSurfer3: Improved Structure-Based Virtual Screening for Structure Variation Using 3D Zernike Descriptors

Structure-based virtual screening (SBVS) is a widely used method in silico drug discovery. It necessitates a receptor structure or binding site to predict the binding pose and fitness of a ligand. Therefore, the performance of the SBVS is affected by the protein conformation. The most frequently used method in SBVS is the protein-ligand docking program, which utilizes atomic distance-based scoring functions. Hence, they are highly prone to sensitivity towards variation in receptor structure, and it is reported that the conformational change significantly drops the performance of the docking program. To address the problem, we have introduced a novel program of SBVS, named PL-PatchSurfer. This program makes use of molecular surface patches and the Zernike descriptor. The surfaces of the pocket and ligand are segmented into several patches by the program. These patches are then mapped with physico-chemical properties such as shape and electrostatic potential before being converted into the Zernike descriptor, which is rotationally invariant. A complementarity between the protein and the ligand is assessed by comparing the descriptors and geometric distribution of the patches in the molecules. A benchmarking study showed that PL-PatchSurfer2 was able to screen active molecules regardless of the receptor structure change with fast speed. However, the program could not achieve high performance for the targets that the hydrogen bonding feature is important such as nuclear hormone receptors. In this paper, we present the newer version of PL-PatchSurfer, PL-PatchSurfer3, which incorporates two new features: a change in the definition of hydrogen bond complementarity and consideration of visibility that contains curvature information of a patch. Our evaluation demonstrates that the new program outperforms its predecessor and other SBVS methods while retaining its characteristic tolerance to receptor structure changes. Interested individuals can access the program at kiharalab.org/plps3.

Improving the Accuracy of Physics-Based Hydration-Free Energy Predictions by Machine Learning the Remaining Error Relative to the Experiment

The accuracy of computational models of water is key to atomistic simulations of biomolecules. We propose a computationally efficient way to improve the accuracy of the prediction of hydration-free energies (HFEs) of small molecules: the remaining errors of the physics-based models relative to the experiment are predicted and mitigated by machine learning (ML) as a postprocessing step. Specifically, the trained graph convolutional neural network attempts to identify the "blind spots" in the physics-based model predictions, where the complex physics of aqueous solvation is poorly accounted for, and partially corrects for them. The strategy is explored for five classical solvent models representing various accuracy/speed trade-offs, from the fast analytical generalized Born (GB) to the popular TIP3P explicit solvent model; experimental HFEs of small neutral molecules from the FreeSolv set are used for the training and testing. For all of the models, the ML correction reduces the resulting root-mean-square error relative to the experiment for HFEs of small molecules, without significant overfitting and with negligible computational overhead. For example, on the test set, the relative accuracy improvement is 47% for the fast analytical GB, making it, after the ML correction, almost as accurate as uncorrected TIP3P. For the TIP3P model, the accuracy improvement is about 39%, bringing the ML-corrected model's accuracy below the 1 kcal/mol threshold. In general, the relative benefit of the ML corrections is smaller for more accurate physics-based models, reaching the lower limit of about 20% relative accuracy gain compared with that of the physics-based treatment alone. The proposed strategy of using ML to learn the remaining error of physics-based models offers a distinct advantage over training ML alone directly on reference HFEs: it preserves the correct overall trend, even well outside of the training set.

Target Prediction by Multiple Virtual Screenings: Analyzing the SARS-CoV-2 Phenotypic Screening by the Docking Simulations Submitted to the MEDIATE Initiative

Phenotypic screenings are usually combined with deconvolution techniques to characterize the mechanism of action for the retrieved hits. These studies can be supported by various computational analyses, although docking simulations are rarely employed. The present study aims to assess if multiple docking calculations can prove successful in target prediction. In detail, the docking simulations submitted to the MEDIATE initiative are utilized to predict the viral targets involved in the hits retrieved by a recently published cytopathic screening. Multiple docking results are combined by the EFO approach to develop target-specific consensus models. The combination of multiple docking simulations enhances the performances of the developed consensus models (average increases in EF1% value of 40% and 25% when combining three and two docking runs, respectively). These models are able to propose reliable targets for about half of the retrieved hits (31 out of 59). Thus, the study emphasizes that docking simulations might be effective in target identification and provide a convincing validation for the collaborative strategies that inspire the MEDIATE initiative. Disappointingly, cross-target and cross-program correlations suggest that common scoring functions are not specific enough for the simulated target.

Ensemble-based virtual screening of African natural products to target human thymidylate synthase

Human thymidylate synthase (hTS) is a validated drug target for chemotherapy. A virtual screening experiment was used to prioritize a list of compounds from African Natural Products Databases docked against the orthosteric binding pocket of hTS. Consensus scores of binding affinities from ensemble-based virtual screening, hydrated docking and MM-PBSA calculations ranked compounds NEA4433 and NEA4434 as the best candidates owing to binding affinity scores in the picomolar order, their excellent ADMET profiles and the good stability of the protein-ligand complexes formed. The current study demonstrates the role of water in small molecule binding to hTS in mediating protein-ligand interactions. Similarly, the robust ensemble docking (relaxed scheme complex) ranked NEA4433 and NEA4434 as the best candidates. Furthermore, the best candidates prioritized were shown to strongly interact with the same residues that interacted with hTS substrate and cofactor.

Integrated Molecular Modeling and Machine Learning for Drug Design

Modern therapeutic development often involves several stages that are interconnected, and multiple iterations are usually required to bring a new drug to the market. Computational approaches have increasingly become an indispensable part of helping reduce the time and cost of the research and development of new drugs. In this Perspective, we summarize our recent efforts on integrating molecular modeling and machine learning to develop computational tools for modulator design, including a pocket-guided rational design approach based on AlphaSpace to target protein-protein interactions, delta machine learning scoring functions for protein-ligand docking as well as virtual screening, and state-of-the-art deep learning models to predict calculated and experimental molecular properties based on molecular mechanics optimized geometries. Meanwhile, we discuss remaining challenges and promising directions for further development and use a retrospective example of FDA approved kinase inhibitor Erlotinib to demonstrate the use of these newly developed computational tools.

Worth the Weight: Sub-Pocket EXplorer (SubPEx), a Weighted Ensemble Method to Enhance Binding-Pocket Conformational Sampling

Structure-based virtual screening (VS) is an effective method for identifying potential small-molecule ligands, but traditional VS approaches consider only a single binding-pocket conformation. Consequently, they struggle to identify ligands that bind to alternate conformations. Ensemble docking helps address this issue by incorporating multiple conformations into the docking process, but it depends on methods that can thoroughly explore pocket flexibility. We here introduce Sub-Pocket EXplorer (SubPEx), an approach that uses weighted ensemble (WE) path sampling to accelerate binding-pocket sampling. As proof of principle, we apply SubPEx to three proteins relevant to drug discovery: heat shock protein 90, influenza neuraminidase, and yeast hexokinase 2. SubPEx is available free of charge without registration under the terms of the open-source MIT license: http://durrantlab.com/subpex/.

Computational insights into the role of structurally diverse plant secondary metabolites as inhibitors against Imidazole Glycerol Phosphate Dehydratase of Mycobacterium tuberculosis

Mycobacterium tuberculosis (Mtb) is one of the major causes of death worldwide and there is a pressing need for the development of novel drug leads. The Imidazole Glycerol Phosphate Dehydratase (IGPD) of Mtb is one of the key enzymes in the histidine biosynthesis pathway and has been recognized as the potentially underexploited drug target for anti-tuberculosis treatment. In the present study, 6063 structurally diverse plant secondary metabolites (PSM) were screened for their efficiency in inhibiting the catalytic activity of IGPD through molecular docking. The top 150 PSMs with the lowest binding energy represent the chemical classes, including Tannins (34%), Flavonoid Glycosides (14%), Terpene Glycosides (10%), Steroid Lactones (9.3%), Flavonoids (6.6%), Steroidal Glycosides (4.6%), etc. Bismahanine, Ashwagandhanolide, and Daurisoline form stable IGPD-inhibitor complexes with binding free energies of -291.3 ± 16.5, -279.0 ± 25.0, and -279.8 ± 17.6 KJ/mol, respectively, as determined by molecular dynamics simulations. These PSM demonstrated strong H-bond interactions with the amino acid residues Ile279, Arg281, and Lys276 in the catalytic region of IGPD, as revealed by structural snapshots. On the basis of our findings, these three PSM could be considered as possible leads against IGPD and should be explored in vitro and in vivo.Communicated by Ramaswamy H. Sarma.

Open-Source Machine Learning in Computational Chemistry

The field of computational chemistry has seen a significant increase in the integration of machine learning concepts and algorithms. In this Perspective, we surveyed 179 open-source software projects, with corresponding peer-reviewed papers published within the last 5 years, to better understand the topics within the field being investigated by machine learning approaches. For each project, we provide a short description, the link to the code, the accompanying license type, and whether the training data and resulting models are made publicly available. Based on those deposited in GitHub repositories, the most popular employed Python libraries are identified. We hope that this survey will serve as a resource to learn about machine learning or specific architectures thereof by identifying accessible codes with accompanying papers on a topic basis. To this end, we also include computational chemistry open-source software for generating training data and fundamental Python libraries for machine learning. Based on our observations and considering the three pillars of collaborative machine learning work, open data, open source (code), and open models, we provide some suggestions to the community.

Worth the weight: Sub-Pocket EXplorer (SubPEx), a weighted-ensemble method to enhance binding-pocket conformational sampling

Structure-based virtual screening (VS) is an effective method for identifying potential small-molecule ligands, but traditional VS approaches consider only a single binding-pocket conformation. Consequently, they struggle to identify ligands that bind to alternate conformations. Ensemble docking helps address this issue by incorporating multiple conformations into the docking process, but it depends on methods that can thoroughly explore pocket flexibility. We here introduce Sub-Pocket EXplorer (SubPEx), an approach that uses weighted ensemble (WE) path sampling to accelerate binding-pocket sampling. As proof of principle, we apply SubPEx to three proteins relevant to drug discovery: heat shock protein 90, influenza neuraminidase, and yeast hexokinase 2. SubPEx is available free of charge without registration under the terms of the open-source MIT license: http://durrantlab.com/subpex/.

Designing Collagen-Binding Peptide with Enhanced Properties Using Hydropathic Free Energy Predictions

Collagen is fundamental to a vast diversity of health functions and potential therapeutics. Short peptides targeting collagen are attractive for designing modular systems for site-specific delivery of bioactive agents. Characterization of peptide-protein binding involves a larger number of potential interactions that require screening methods to target physiological conditions. We build a hydropathy-based free energy estimation tool which allows quick evaluation of peptides binding to collagen. Previous studies showed that pH plays a significant role in collagen structure and stability. Our design tool enables probing peptides for their collagen-binding property across multiple pH conditions. We explored binding features of currently known collagen-binding peptides, collagen type I alpha chain 2 sense peptide (TKKTLRT) and decorin LRR-10 (LRELHLNNN). Based on these analyzes, we engineered a collagen-binding peptide with enhanced properties across a large pH range in contrast to LRR-10 pH dependence. To validate our predictions, we used a quantum-dots-based binding assay to compare the coverage of the peptides on type I collagen. The predicted peptide resulted in improved collagen binding. Hydropathy of the peptide-protein pair is a promising approach to finding compatible pairings with minimal use of computational resources, and our method allows for quick evaluation of peptides for binding to other proteins. Overall, the free-energy-based tool provides an alternative computational screening approach that impacts protein interaction search methods.

Advances in the discovery of new chemotypes through ultra-large library docking

INTRODUCTION

The size and complexity of virtual screening libraries in drug discovery have skyrocketed in recent years, reaching up to multiple billions of accessible compounds. However, virtual screening of such ultra-large libraries poses several challenges associated with preparing the libraries, sampling, and pre-selection of suitable compounds. The utilization of artificial intelligence (AI)-assisted screening approaches, such as deep learning, poses a promising countermeasure to deal with this rapidly expanding chemical space. For example, various AI-driven methods were recently successfully used to identify novel small molecule inhibitors of the SARS-CoV-2 main protease (M^pro).

AREAS COVERED

This review focuses on presenting various kinds of virtual screening methods suitable for dealing with ultra-large libraries. Challenges associated with these computational methodologies are discussed, and recent advances are highlighted in the example of the discovery of novel M^pro inhibitors targeting the SARS-CoV-2 virus.

EXPERT OPINION

With the rapid expansion of the virtual chemical space, the methodologies for docking and screening such quantities of molecules need to keep pace. Employment of AI-driven screening compounds has already been shown to be effective in a range from a few thousand to multiple billion compounds, furthered by de novo generation of drug-like molecules without human interference.

Machine Learning in Metastatic Cancer Research: Potentials, Possibilities, and Prospects

Cancer has received extensive recognition for its high mortality rate, with metastatic cancer being the top cause of cancer-related deaths. Metastatic cancer involves the spread of the primary tumor to other body organs. As much as the early detection of cancer is essential, the timely detection of metastasis, the identification of biomarkers, and treatment choice are valuable for improving the quality of life for metastatic cancer patients. This study reviews the existing studies on classical machine learning (ML) and deep learning (DL) in metastatic cancer research. Since the majority of metastatic cancer research data are collected in the formats of PET/CT and MRI image data, deep learning techniques are heavily involved. However, its black-box nature and expensive computational cost are notable concerns. Furthermore, existing models could be overestimated for their generality due to the non-diverse population in clinical trial datasets. Therefore, research gaps are itemized; follow-up studies should be carried out on metastatic cancer using machine learning and deep learning tools with data in a symmetric manner.

Computational Chemistry for the Identification of Lead Compounds for Radiotracer Development

The use of computer-aided drug design (CADD) for the identification of lead compounds in radiotracer development is steadily increasing. Traditional CADD methods, such as structure-based and ligand-based virtual screening and optimization, have been successfully utilized in many drug discovery programs and are highlighted throughout this review. First, we discuss the use of virtual screening for hit identification at the beginning of drug discovery programs. This is followed by an analysis of how the hits derived from virtual screening can be filtered and culled to highly probable candidates to test in in vitro assays. We then illustrate how CADD can be used to optimize the potency of experimentally validated hit compounds from virtual screening for use in positron emission tomography (PET). Finally, we conclude with a survey of the newest techniques in CADD employing machine learning (ML).

AlphaFold2 protein structure prediction: Implications for drug discovery

The drug discovery process involves designing compounds to selectively interact with their targets. The majority of therapeutic targets for low molecular weight (small molecule) drugs are proteins. The outstanding accuracy with which recent artificial intelligence methods compile the three-dimensional structure of proteins has made protein targets more accessible to the drug design process. Here, we present our perspective of the significance of accurate protein structure prediction on various stages of the small molecule drug discovery life cycle focusing on current capabilities and assessing how further evolution of such predictive procedures can have a more decisive impact in the discovery of new medicines.

Targeting human thymidylate synthase: Ensemble-based virtual screening for drug repositioning and the role of water

A drug repositioning computational approach was carried to search inhibitors for human thymidylate synthase. An ensemble-based virtual screening of FDA-approved drugs showed the drugs Imatinib, Lumacaftor and Naldemedine to be likely candidates for repurposing. The role of water in the drug-receptor interactions was revealed by the application of an extended AutoDock scoring function that included the water forcefield. The binding affinity scores when hydrated ligands were docked were improved in the drugs considered. Further binding free energy calculations based on the Molecular Mechanics Poisson-Boltzmann Surface Area method revealed that Imatinib, Lumacaftor and Naldemedine scored -130.7 ± 28.1, -210.6 ± 29.9 and -238.0 ± 25.4 kJ/mol, respectively, showing good binding affinity for the candidates considered. Overall, the analysis of the molecular dynamics trajectory of the receptor-drug complexes revealed stable structures for Imatinib, Lumacaftor and Naldemedine, for the entire simulation time.

Machine learning driven drug repurposing strategy for identification of potential RET inhibitors against non-small cell lung cancer

Non-small cell lung cancer (NSCLC) remains the leading cause of mortality and morbidity worldwide accounting about 85% of total lung cancer cases. The receptor REarranged during Transfection (RET) plays an important role by ligand independent activation of kinase domain resulting in carcinogenesis. Presently, the treatment for RET driven NSCLC is limited to multiple kinase inhibitors. This situation necessitates the discovery of novel and potent RET specific inhibitors. Thus, we employed high throughput screening strategy to repurpose FDA approved compounds from DrugBank comprising of 2509 molecules. It is worth noting that the initial screening is accomplished with the aid of in-house machine learning model built using IC₅₀ values corresponding to 2854 compounds obtained from BindingDB repository. A total of 497 compounds (19%) were predicted as actives by our generated model. Subsequent in silico validation process such as molecular docking, MMGBSA and density function theory analysis resulted in identification of two lead compounds named DB09313 and DB00471. The simulation study highlights the potency of DB00471 (Montelukast) as potential RET inhibitor among the investigated compounds. In the end, the half-minimal inhibitory activity of montelukast was also predicted against RET protein expressing LC-2/ad cell lines demonstrated significant anticancer activity. Collective analysis from our study highlights that montelukast could be a promising candidate for the management of RET specific NSCLC.

A Guide to In Silico Drug Design

The drug discovery process is a rocky path that is full of challenges, with the result that very few candidates progress from hit compound to a commercially available product, often due to factors, such as poor binding affinity, off-target effects, or physicochemical properties, such as solubility or stability. This process is further complicated by high research and development costs and time requirements. It is thus important to optimise every step of the process in order to maximise the chances of success. As a result of the recent advancements in computer power and technology, computer-aided drug design (CADD) has become an integral part of modern drug discovery to guide and accelerate the process. In this review, we present an overview of the important CADD methods and applications, such as in silico structure prediction, refinement, modelling and target validation, that are commonly used in this area.

Finding alternatives to 5-fluorouracil: application of ensemble-based virtual screening for drug repositioning against human thymidylate synthase

5-fluorouracil and analogs are used in the treatment of many solid tumours. However, there are many cases of resistance and high toxicity associated with 5-fluorouracil chemotherapy. Repurposing FDA drugs against human thymidylate synthase revealed a number of FDA drugs that have a potential to be further developed for the treatment of various cancers for which 5-fluorouracil and analogs have been used for chemotherapy. Four FDA drugs prioritized for further validation included Erismodegib, Irinotecan, Conivaptan and Ergotamine. The role of water in mediating drug interactions and its contribution to the total binding energy was also shown. MM-PBSA calculations revealed that the binding affinity was the lowest for the hTS-Ergotamine complex (-66.702 ± 1.807 kJ/mol) suggesting moderate inhibition despite a large energetic contribution from van der Waal interactions (-190.889 ± 1.027 kJ/mol).Communicated by Ramaswamy H. Sarma.

Collaborative Approach between Explainable Artificial Intelligence and Simplified Chemical Interactions to Explore Active Ligands for Cyclin-Dependent Kinase 2

To improve virtual screening for drug discovery, we present a collaborative approach between explainable artificial intelligence (AI) and simplified chemical interaction scores to efficiently search for active ligands bound to the target receptor. In particular, we focus on cyclin-dependent kinase 2 (CDK2), which is well known as a cancer target protein. Docking simulation alone is insufficient to distinguish active ligands from decoy molecules. To identify active ligands, in this paper, machine learning is employed together with scoring functions that simplify the screened Coulomb and Lennard-Jones interactions between the ligands and residues of the target receptor. We demonstrate that these simplified interaction scores can significantly improve the classification ability of machine learning models. We also demonstrate that explainable AI together with the simplified scoring method can highlight the important residues of CDK2 for recognizing active ligands.

Ensemble learning from ensemble docking: revisiting the optimum ensemble size problem

Despite considerable advances obtained by applying machine learning approaches in protein–ligand affinity predictions, the incorporation of receptor flexibility has remained an important bottleneck. While ensemble docking has been used widely as a solution to this problem, the optimum choice of receptor conformations is still an open question considering the issues related to the computational cost and false positive pose predictions. Here, a combination of ensemble learning and ensemble docking is suggested to rank different conformations of the target protein in light of their importance for the final accuracy of the model. Available X-ray structures of cyclin-dependent kinase 2 (CDK2) in complex with different ligands are used as an initial receptor ensemble, and its redundancy is removed through a graph-based redundancy removal, which is shown to be more efficient and less subjective than clustering-based representative selection methods. A set of ligands with available experimental affinity are docked to this nonredundant receptor ensemble, and the energetic features of the best scored poses are used in an ensemble learning procedure based on the random forest method. The importance of receptors is obtained through feature selection measures, and it is shown that a few of the most important conformations are sufficient to reach 1 kcal/mol accuracy in affinity prediction with considerable improvement of the early enrichment power of the models compared to the different ensemble docking without learning strategies. A clear strategy has been provided in which machine learning selects the most important experimental conformers of the receptor among a large set of protein–ligand complexes while simultaneously maintaining the final accuracy of affinity predictions at the highest level possible for available data. Our results could be informative for future attempts to design receptor-specific docking-rescoring strategies.

Virtual Screening with Gnina 1.0

Virtual screening-predicting which compounds within a specified compound library bind to a target molecule, typically a protein-is a fundamental task in the field of drug discovery. Doing virtual screening well provides tangible practical benefits, including reduced drug development costs, faster time to therapeutic viability, and fewer unforeseen side effects. As with most applied computational tasks, the algorithms currently used to perform virtual screening feature inherent tradeoffs between speed and accuracy. Furthermore, even theoretically rigorous, computationally intensive methods may fail to account for important effects relevant to whether a given compound will ultimately be usable as a drug. Here we investigate the virtual screening performance of the recently released Gnina molecular docking software, which uses deep convolutional networks to score protein-ligand structures. We find, on average, that Gnina outperforms conventional empirical scoring. The default scoring in Gnina outperforms the empirical AutoDock Vina scoring function on 89 of the 117 targets of the DUD-E and LIT-PCBA virtual screening benchmarks with a median 1% early enrichment factor that is more than twice that of Vina. However, we also find that issues of bias linger in these sets, even when not used directly to train models, and this bias obfuscates to what extent machine learning models are achieving their performance through a sophisticated interpretation of molecular interactions versus fitting to non-informative simplistic property distributions.