CSAR 2014: A Benchmark Exercise Using Unpublished Data from Pharma

Machine Learning for Sequence and Structure-Based Protein-Ligand Interaction Prediction

Developing new drugs is too expensive and time -consuming. Accurately predicting the interaction between drugs and targets will likely change how the drug is discovered. Machine learning-based protein-ligand interaction prediction has demonstrated significant potential. In this paper, computational methods, focusing on sequence and structure to study protein-ligand interactions, are examined. Therefore, this paper starts by presenting an overview of the data sets applied in this area, as well as the various approaches applied for representing proteins and ligands. Then, sequence-based and structure-based classification criteria are subsequently utilized to categorize and summarize both the classical machine learning models and deep learning models employed in protein-ligand interaction studies. Moreover, the evaluation methods and interpretability of these models are proposed. Furthermore, delving into the diverse applications of protein-ligand interaction models in drug research is presented. Lastly, the current challenges and future directions in this field are addressed.

Automated benchmarking of combined protein structure and ligand conformation prediction

The prediction of protein-ligand complexes (PLC), using both experimental and predicted structures, is an active and important area of research, underscored by the inclusion of the Protein-Ligand Interaction category in the latest round of the Critical Assessment of Protein Structure Prediction experiment CASP15. The prediction task in CASP15 consisted of predicting both the three-dimensional structure of the receptor protein as well as the position and conformation of the ligand. This paper addresses the challenges and proposed solutions for devising automated benchmarking techniques for PLC prediction. The reliability of experimentally solved PLC as ground truth reference structures is assessed using various validation criteria. Similarity of PLC to previously released complexes are employed to judge PLC diversity and the difficulty of a PLC as a prediction target. We show that the commonly used PDBBind time-split test-set is inappropriate for comprehensive PLC evaluation, with state-of-the-art tools showing conflicting results on a more representative and high quality dataset constructed for benchmarking purposes. We also show that redocking on crystal structures is a much simpler task than docking into predicted protein models, demonstrated by the two PLC-prediction-specific scoring metrics created. Finally, we introduce a fully automated pipeline that predicts PLC and evaluates the accuracy of the protein structure, ligand pose, and protein-ligand interactions.

Template-guided method for protein-ligand complex structure prediction: Application to CASP15 protein-ligand studies

Critical Assessment of Structure Prediction 15 (CASP15) added a new category of ligand prediction to promote the development of protein/RNA-ligand modeling methods, which have become important tools in modern drug discovery. A total of 22 targets were released, including 18 protein-ligand targets and 4 RNA-ligand targets. We applied our recently developed template-guided method to the protein-ligand complex structure predictions. The method combined a physicochemical, molecular docking method, and a bioinformatics-based ligand similarity method. The Protein Data Bank was scanned for template structures containing the target protein, homologous proteins, or proteins sharing a similar fold with the target protein. The binding modes of the co-bound ligands in the template structures were used to guide the complex structure prediction for the target. The CASP assessment results show that the overall performance of our method was ranked second when the top predicted model was considered for each target. Here, we analyzed our predictions in detail, and discussed the challenges including protein conformational changes, large and flexible ligands, and multiple diverse ligands in a binding pocket.

Evaluating the druggability of TrmD, a potential antibacterial target, through design and microbiological profiling of a series of potent TrmD inhibitors

The post-transcriptional modifier tRNA-(N¹G37) methyltransferase (TrmD) has been proposed to be essential for growth in many Gram-negative and Gram-positive pathogens, however previously reported inhibitors show only weak antibacterial activity. In this work, optimisation of fragment hits resulted in compounds with low nanomolar TrmD inhibition incorporating features designed to enhance bacterial permeability and covering a range of physicochemical space. The resulting lack of significant antibacterial activity suggests that whilst TrmD is highly ligandable, its essentiality and druggability are called into question.

Antifungal Activity, Antibiofilm and Association Studies with O-Alkylamidoximes against Cryptococcus spp

This is the first study that describes the antifungal and anti-biofilm potential of O-alkylamidoximes against strains of Cryptococcus neoformans and Cryptococcus gattii. In vitro tests have shown that O-alkylamidoximes are capable of inhibiting fungal growth and biofilm formation of the C. neoformans and C. gattii strains, suggesting, from molecular docking, the potential for interaction with the Hsp90. The associations between O-alkylamidoximes and amphotericin B were beneficial. Therefore, O-alkylamidoximes can be a useful alternative to contribute to the limited arsenal of drugs, since they showed a powerful action against the primary agents of Cryptococcosis.

Recent PELE Developments and Applications in Drug Discovery Campaigns

Computer simulation techniques are gaining a central role in molecular pharmacology. Due to several factors, including the significant improvements of traditional molecular modelling, the irruption of machine learning methods, the massive data generation, or the unlimited computational resources through cloud computing, the future of pharmacology seems to go hand in hand with in silico predictions. In this review, we summarize our recent efforts in such a direction, centered on the unconventional Monte Carlo PELE software and on its coupling with machine learning techniques. We also provide new data on combining two recent new techniques, aquaPELE capable of exhaustive water sampling and fragPELE, for fragment growing.

SCORCH: Improving structure-based virtual screening with machine learning classifiers, data augmentation, and uncertainty estimation

INTRODUCTION

The discovery of a new drug is a costly and lengthy endeavour. The computational prediction of which small molecules can bind to a protein target can accelerate this process if the predictions are fast and accurate enough. Recent machine-learning scoring functions re-evaluate the output of molecular docking to achieve more accurate predictions. However, previous scoring functions were trained on crystalised protein-ligand complexes and datasets of decoys. The limited availability of crystal structures and biases in the decoy datasets can lower the performance of scoring functions.

OBJECTIVES

To address key limitations of previous scoring functions and thus improve the predictive performance of structure-based virtual screening.

METHODS

A novel machine-learning scoring function was created, named SCORCH (Scoring COnsensus for RMSD-based Classification of Hits). To develop SCORCH, training data is augmented by considering multiple ligand poses and labelling poses based on their RMSD from the native pose. Decoy bias is addressed by generating property-matched decoys for each ligand and using the same methodology for preparing and docking decoys and ligands. A consensus of 3 different machine learning approaches is also used to improve performance.

RESULTS

We find that multi-pose augmentation in SCORCH improves its docking power and screening power on independent benchmark datasets. SCORCH outperforms an equivalent scoring function trained on single poses, with a 1% enrichment factor (EF) of 13.78 vs. 10.86 on 18 DEKOIS 2.0 targets and a mean native pose rank of 5.9 vs 30.4 on CSAR 2014. Additionally, SCORCH outperforms widely used scoring functions in virtual screening and pose prediction on independent benchmark datasets.

CONCLUSION

By rationally addressing key limitations of previous scoring functions, SCORCH improves the performance of virtual screening. SCORCH also provides an estimate of its uncertainty, which can help reduce the cost and time required for drug discovery.

Protein-Ligand Docking in the Machine-Learning Era

Molecular docking plays a significant role in early-stage drug discovery, from structure-based virtual screening (VS) to hit-to-lead optimization, and its capability and predictive power is critically dependent on the protein-ligand scoring function. In this review, we give a broad overview of recent scoring function development, as well as the docking-based applications in drug discovery. We outline the strategies and resources available for structure-based VS and discuss the assessment and development of classical and machine learning protein-ligand scoring functions. In particular, we highlight the recent progress of machine learning scoring function ranging from descriptor-based models to deep learning approaches. We also discuss the general workflow and docking protocols of structure-based VS, such as structure preparation, binding site detection, docking strategies, and post-docking filter/re-scoring, as well as a case study on the large-scale docking-based VS test on the LIT-PCBA data set.

Scoring Functions for Protein-Ligand Binding Affinity Prediction using Structure-Based Deep Learning: A Review

The rapid and accurate in silico prediction of protein-ligand binding free energies or binding affinities has the potential to transform drug discovery. In recent years, there has been a rapid growth of interest in deep learning methods for the prediction of protein-ligand binding affinities based on the structural information of protein-ligand complexes. These structure-based scoring functions often obtain better results than classical scoring functions when applied within their applicability domain. Here we review structure-based scoring functions for binding affinity prediction based on deep learning, focussing on different types of architectures, featurization strategies, data sets, methods for training and evaluation, and the role of explainable artificial intelligence in building useful models for real drug-discovery applications.

PlaceWaters: Real-time, explicit interface water sampling during Rosetta ligand docking

Water molecules at the protein-small molecule interface often form hydrogen bonds with both the small molecule ligand and the protein, affecting the structural integrity and energetics of a binding event. The inclusion of these 'bridging waters' has been shown to improve the accuracy of predicted docked structures; however, due to increased computational costs, this step is typically omitted in ligand docking simulations. In this study, we introduce a resource-efficient, Rosetta-based protocol named "PlaceWaters" to predict the location of explicit interface bridging waters during a ligand docking simulation. In contrast to other explicit water methods, this protocol is independent of knowledge of number and location of crystallographic waters in homologous structures. We test this method on a diverse protein-small molecule benchmark set in comparison to other Rosetta-based protocols. Our results suggest that this coarse-grained, structure-based approach quickly and accurately predicts the location of bridging waters, improving our ability to computationally screen drug candidates.

Design of Artificial Enzymes Bearing Several Active Centers: New Trends, Opportunities and Problems

Harnessing enzymes which possess several catalytic activities is a topic where intense research has been carried out, mainly coupled with the development of cascade reactions. This review tries to cover the different possibilities to reach this goal: enzymes with promiscuous activities, fusion enzymes, enzymes + metal catalysts (including metal nanoparticles or site-directed attached organometallic catalyst), enzymes bearing non-canonical amino acids + metal catalysts, design of enzymes bearing a second biological but artificial active center (plurizymes) by coupling enzyme modelling and directed mutagenesis and plurizymes that have been site directed modified in both or in just one active center with an irreversible inhibitor attached to an organometallic catalyst. Some examples of cascade reactions catalyzed by the enzymes bearing several catalytic activities are also described. Finally, some foreseen problems of the use of these multi-activity enzymes are described (mainly related to the balance of the catalytic activities, necessary in many instances, or the different operational stabilities of the different catalytic activities). The design of new multi-activity enzymes (e.g., plurizymes or modified plurizymes) seems to be a topic with unarguable interest, as this may link biological and non-biological activities to establish new combo-catalysis routes.

Advances in the molecular understanding of G protein-coupled receptors and their future therapeutic opportunities

Background

Understanding the mechanisms, activated and inhibited pathways as well as other molecular targets involved in existing and emerging disease conditions provides useful insights into their proper diagnosis and treatment and aids drug discovery, development and production. G protein-coupled receptors (GPCRs) are one of the most important classes of targets for small-molecule drug discovery. Of all drug targets, GPCRs are the most studied, undoubtedly because of their pharmacological tractability and role in the pathophysiology as well as the pathogenesis of human diseases.

Main body of the abstract

GPCRs are regarded as the largest target class of the “druggable genome” representing approximately 19% of the currently available drug targets. They have long played a prominent role in drug discovery, such that as of this writing, 481 drugs (about 34% of all FDA-approved drugs) act on GPCRs. More than 320 therapeutic agents are currently under clinical trials, of which a significant percentage targets novel GPCRs. GPCRs are implicated in a wide variety of diseases including CNS disorders, inflammatory diseases such as rheumatoid arthritis and Crohn’s disease, as well as metabolic disease and cancer. The non-olfactory human GPCRs yet to be clinically explored or tried are endowed with perhaps a huge untapped potential drug discovery especially in the field of immunology and genetics.

Short conclusion

This review discusses the recent advances in the molecular pharmacology and future opportunities for targeting GPCRs with a view to drug development.

KORP-PL: a coarse-grained knowledge-based scoring function for protein-ligand interactions

MOTIVATION

Despite the progress made in studying protein-ligand interactions and the widespread application of docking and affinity prediction tools, improving their precision and efficiency still remains a challenge. Computational approaches based on the scoring of docking conformations with statistical potentials constitute a popular alternative to more accurate but costly physics-based thermodynamic sampling methods. In this context, a minimalist and fast sidechain-free knowledge-based potential with a high docking and screening power can be very useful when screening a big number of putative docking conformations.

RESULTS

Here, we present a novel coarse-grained potential defined by a 3D joint probability distribution function that only depends on the pairwise orientation and position between protein backbone and ligand atoms. Despite its extreme simplicity, our approach yields very competitive results with the state-of-the-art scoring functions, especially in docking and screening tasks. For example, we observed a twofold improvement in the median 5% enrichment factor on the DUD-E benchmark compared to Autodock Vina results. Moreover, our results prove that a coarse sidechain-free potential is sufficient for a very successful docking pose prediction.

AVAILABILITYAND IMPLEMENTATION

The standalone version of KORP-PL with the corresponding tests and benchmarks are available at https://team.inria.fr/nano-d/korp-pl/ and https://chaconlab.org/modeling/korp-pl.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Identification of Cryptic Binding Sites Using MixMD with Standard and Accelerated Molecular Dynamics

Protein dynamics play an important role in small molecule binding and can pose a significant challenge in the identification of potential binding sites. Cryptic binding sites have been defined as sites which require significant rearrangement of the protein structure to become physically accessible to a ligand. Mixed-solvent MD (MixMD) is a computational protocol which maps the surface of the protein using molecular dynamics (MD) of the unbound protein solvated in a 5% box of probe molecules with explicit water. This method has successfully identified known active and allosteric sites which did not require reorganization. In this study, we apply the MixMD protocol to identify known cryptic sites of 12 proteins characterized by a wide range of conformational changes. Of these 12 proteins, three require reorganization of side chains, five require loop movements, and four require movement of more significant structures such as whole helices. In five cases, we find that standard MixMD simulations are able to map the cryptic binding sites with at least one probe type. In two cases (guanylate kinase and TIE-2), accelerated MD, which increases sampling of torsional angles, was necessary to achieve mapping of portions of the cryptic binding site missed by standard MixMD. For more complex systems where movement of a helix or domain is necessary, MixMD was unable to map the binding site even with accelerated dynamics, possibly due to the limited timescale (100 ns for individual simulations). In general, similar conformational dynamics are observed in water-only simulations and those with probe molecules. This could imply that the probes are not driving opening events but rather take advantage of mapping sites that spontaneously open as part of their inherent conformational behavior. Finally, we show that docking to an ensemble of conformations from the standard MixMD simulations performs better than docking the apo crystal structure in nine cases and even better than half of the bound crystal structures. Poorer performance was seen in docking to ensembles of conformations from the accelerated MixMD simulations.

Computational design of enzymes for biotechnological applications

Enzymes are the natural catalysts that execute biochemical reactions upholding life. Their natural effectiveness has been fine-tuned as a result of millions of years of natural evolution. Such catalytic effectiveness has prompted the use of biocatalysts from multiple sources on different applications, including the industrial production of goods (food and beverages, detergents, textile, and pharmaceutics), environmental protection, and biomedical applications. Natural enzymes often need to be improved by protein engineering to optimize their function in non-native environments. Recent technological advances have greatly facilitated this process by providing the experimental approaches of directed evolution or by enabling computer-assisted applications. Directed evolution mimics the natural selection process in a highly accelerated fashion at the expense of arduous laboratory work and economic resources. Theoretical methods provide predictions and represent an attractive complement to such experiments by waiving their inherent costs. Computational techniques can be used to engineer enzymatic reactivity, substrate specificity and ligand binding, access pathways and ligand transport, and global properties like protein stability, solubility, and flexibility. Theoretical approaches can also identify hotspots on the protein sequence for mutagenesis and predict suitable alternatives for selected positions with expected outcomes. This review covers the latest advances in computational methods for enzyme engineering and presents many successful case studies.

Artificial Intelligence in Drug Discovery: A Comprehensive Review of Data-driven and Machine Learning Approaches

As expenditure on drug development increases exponentially, the overall drug discovery process requires a sustainable revolution. Since artificial intelligence (AI) is leading the fourth industrial revolution, AI can be considered as a viable solution for unstable drug research and development. Generally, AI is applied to fields with sufficient data such as computer vision and natural language processing, but there are many efforts to revolutionize the existing drug discovery process by applying AI. This review provides a comprehensive, organized summary of the recent research trends in AI-guided drug discovery process including target identification, hit identification, ADMET prediction, lead optimization, and drug repositioning. The main data sources in each field are also summarized in this review. In addition, an in-depth analysis of the remaining challenges and limitations will be provided, and proposals for promising future directions in each of the aforementioned areas.

Biased Docking for Protein-Ligand Pose Prediction

The interaction between a protein and its ligands is one of the basic and most important processes in biological chemistry. Docking methods aim to predict the molecular 3D structure of protein-ligand complexes starting from coordinates of the protein and the ligand separately. They are widely used in both industry and academia, especially in the context of drug development projects. AutoDock4 is one of the most popular docking tools and, as for any docking method, its performance is highly system dependent. Knowledge about specific protein-ligand interactions on a particular target can be used to successfully overcome this limitation. Here, we describe how to apply the AutoDock Bias protocol, a simple and elegant strategy that allows users to incorporate target-specific information through a modified scoring function that biases the ligand structure towards those poses (or conformations) that establish selected interactions. We discuss two examples using different bias sources. In the first, we show how to steer dockings towards interactions derived from crystal structures of the receptor with different ligands; in the second example, we define and apply hydrophobic biases derived from Molecular Dynamics simulations in mixed solvents. Finally, we discuss general concepts of biased docking, its performance in pose prediction, and virtual screening campaigns as well as other potential applications.

RASPD+: Fast Protein-Ligand Binding Free Energy Prediction Using Simplified Physicochemical Features

The virtual screening of large numbers of compounds against target protein binding sites has become an integral component of drug discovery workflows. This screening is often done by computationally docking ligands into a protein binding site of interest, but this has the drawback of a large number of poses that must be evaluated to obtain accurate estimates of protein-ligand binding affinity. We here introduce a fast pre-filtering method for ligand prioritization that is based on a set of machine learning models and uses simple pose-invariant physicochemical descriptors of the ligands and the protein binding pocket. Our method, Rapid Screening with Physicochemical Descriptors + machine learning (RASPD+), is trained on PDBbind data and achieves a regression performance that is better than that of the original RASPD method and traditional scoring functions on a range of different test sets without the need for generating ligand poses. Additionally, we use RASPD+ to identify molecular features important for binding affinity and assess the ability of RASPD+ to enrich active molecules from decoys.

DUBS: A Framework for Developing Directory of Useful Benchmarking Sets for Virtual Screening

Benchmarking is a crucial step in evaluating virtual screening methods for drug discovery. One major issue that arises among benchmarking data sets is a lack of a standardized format for representing the protein and ligand structures used to benchmark the virtual screening method. To address this, we introduce the Directory of Useful Benchmarking Sets (DUBS) framework, as a simple and flexible tool to rapidly create benchmarking sets using the protein databank. DUBS uses a simple input text based format along with the Lemon data mining framework to efficiently access and organize data to the protein databank and output commonly used inputs for virtual screening software. The simple input format used by DUBS allows users to define their own benchmarking data sets and access the corresponding information directly from the software package. Currently, it only takes DUBS less than 2 min to create a benchmark using this format. Since DUBS uses a simple python script, users can easily modify this to create more complex benchmarks. We hope that DUBS will be a useful community resource to provide a standardized representation for benchmarking data sets in virtual screening. The DUBS package is available on GitHub at https://github.com/chopralab/lemon/tree/master/dubs.

Computationally predicting binding affinity in protein-ligand complexes: free energy-based simulations and machine learning-based scoring functions

Accurately predicting protein-ligand binding affinities can substantially facilitate the drug discovery process, but it remains as a difficult problem. To tackle the challenge, many computational methods have been proposed. Among these methods, free energy-based simulations and machine learning-based scoring functions can potentially provide accurate predictions. In this paper, we review these two classes of methods, following a number of thermodynamic cycles for the free energy-based simulations and a feature-representation taxonomy for the machine learning-based scoring functions. More recent deep learning-based predictions, where a hierarchy of feature representations are generally extracted, are also reviewed. Strengths and weaknesses of the two classes of methods, coupled with future directions for improvements, are comparatively discussed.

farPPI: a webserver for accurate prediction of protein-ligand binding structures for small-molecule PPI inhibitors by MM/PB(GB)SA methods

SUMMARY

Protein-protein interactions (PPIs) have been regarded as an attractive emerging class of therapeutic targets for the development of new treatments. Computational approaches, especially molecular docking, have been extensively employed to predict the binding structures of PPI-inhibitors or discover novel small molecule PPI inhibitors. However, due to the relatively 'undruggable' features of PPI interfaces, accurate predictions of the binding structures for ligands towards PPI targets are quite challenging for most docking algorithms. Here, we constructed a non-redundant pose ranking benchmark dataset for small-molecule PPI inhibitors, which contains 900 binding poses for 184 protein-ligand complexes. Then, we evaluated the performance of MM/PB(GB)SA approaches to identify the correct binding poses for PPI inhibitors, including two Prime MM/GBSA procedures from the Schrödinger suite and seven different MM/PB(GB)SA procedures from the Amber package. Our results showed that MM/PBSA outperformed the Glide SP scoring function (success rate of 58.6%) and MM/GBSA in most cases, especially the PB3 procedure which could achieve an overall success rate of ∼74%. Moreover, the GB6 procedure (success rate of 68.9%) performed much better than the other MM/GBSA procedures, highlighting the excellent potential of the GBNSR6 implicit solvation model for pose ranking. Finally, we developed the webserver of Fast Amber Rescoring for PPI Inhibitors (farPPI), which offers a freely available service to rescore the docking poses for PPI inhibitors by using the MM/PB(GB)SA methods.

AVAILABILITY AND IMPLEMENTATION

farPPI web server is freely available at http://cadd.zju.edu.cn/farppi/.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

CANDOCK: Chemical Atomic Network-Based Hierarchical Flexible Docking Algorithm Using Generalized Statistical Potentials

Small-molecule docking has proven to be invaluable for drug design and discovery. However, existing docking methods have several limitations such as improper treatment of the interactions of essential components in the chemical environment of the binding pocket (e.g., cofactors, metal ions, etc.), incomplete sampling of chemically relevant ligand conformational space, and the inability to consistently correlate docking scores of the best binding pose with experimental binding affinities. We present CANDOCK, a novel docking algorithm, that utilizes a hierarchical approach to reconstruct ligands from an atomic grid using graph theory and generalized statistical potential functions to sample biologically relevant ligand conformations. Our algorithm accounts for protein flexibility, solvent, metal ions, and cofactor interactions in the binding pocket that are traditionally ignored by current methods. We evaluate the algorithm on the PDBbind, Astex, and PINC proteins to show its ability to reproduce the binding mode of the ligands that is independent of the initial ligand conformation in these benchmarks. Finally, we identify the best selector and ranker potential functions such that the statistical score of the best selected docked pose correlates with the experimental binding affinities of the ligands for any given protein target. Our results indicate that CANDOCK is a generalized flexible docking method that addresses several limitations of current docking methods by considering all interactions in the chemical environment of a binding pocket for correlating the best-docked pose with biological activity. CANDOCK along with all structures and scripts used for benchmarking is available at https://github.com/chopralab/candock_benchmark.

FragPELE: Dynamic Ligand Growing within a Binding Site. A Novel Tool for Hit-To-Lead Drug Design

The early stages of drug discovery rely on hit-to-lead programs, where initial hits undergo partial optimization to improve binding affinities for their biological target. This is an expensive and time-consuming process, requiring multiple iterations of trial and error designs, an ideal scenario for applying computer simulation. However, most state-of-the-art modeling techniques fail to provide a fast and reliable answer to the Induced-Fit protein-ligand problem. To aid in this matter, we present FragPELE, a new tool for in silico hit-to-lead drug design, capable of growing a fragment from a bound core while exploring the protein-ligand conformational space. We tested the ability of FragPELE to predict crystallographic data, even in cases where cryptic sub-pockets open because of the presence of particular R-groups. Additionally, we evaluated the potential of the software on growing and scoring five congeneric series from the 2015 FEP+ dataset, comparing them to FEP+, SP and Induced-Fit Glide, and MMGBSA simulations. Results show that FragPELE could be useful not only for finding new cavities and novel binding modes in cases where standard docking tools cannot but also to rank ligand activities in a reasonable amount of time and with acceptable precision.

D3R grand challenge 4: blind prediction of protein-ligand poses, affinity rankings, and relative binding free energies

The Drug Design Data Resource (D3R) aims to identify best practice methods for computer aided drug design through blinded ligand pose prediction and affinity challenges. Herein, we report on the results of Grand Challenge 4 (GC4). GC4 focused on proteins beta secretase 1 and Cathepsin S, and was run in an analogous manner to prior challenges. In Stage 1, participant ability to predict the pose and affinity of BACE1 ligands were assessed. Following the completion of Stage 1, all BACE1 co-crystal structures were released, and Stage 2 tested affinity rankings with co-crystal structures. We provide an analysis of the results and discuss insights into determined best practice methods.

GRAM: A True Null Model for Relative Binding Affinity Predictions

Relative binding affinity prediction is a critical component in computer aided drug design. A significant amount of effort has been dedicated to developing rapid and reliable in silico methods. However, robust assessment of their performance is still a complicated issue, as it requires a performance measure applicable in the prospective setting and more importantly a true null model that defines the expected performance of being random in an objective manner. Although many performance metrics, such as the Pearson correlation coefficient (r), mean unsigned error (MUE), and root-mean-square error (RMSE), are frequently used in the literature, a true and nontrivial null model has yet been identified. To address this problem, here we introduce an interval estimate as an additional measure, namely, the prediction interval (PI), which can be estimated from the error distribution of the predictions. The benefits of using the interval estimate are (1) it provides the uncertainty range in the predicted activities, which is important in prospective applications, and (2) a true null model with well-defined PI can be established. We provide one such example termed the Gaussian Random Affinity Model (GRAM), which is based on the empirical observation that the affinity change in a typical lead optimization effort has the tendency to distribute normally N (0, σ). Having an analytically defined PI that only depends on the variation in the activities, GRAM should, in principle, allow us to compare the performance of relative binding affinity prediction methods in a standard way, ultimately critical to measuring the progress made in algorithm development.

Docking rigid macrocycles using Convex-PL, AutoDock Vina, and RDKit in the D3R Grand Challenge 4

The D3R Grand Challenge 4 provided a brilliant opportunity to test macrocyclic docking protocols on a diverse high-quality experimental data. We participated in both pose and affinity prediction exercises. Overall, we aimed to use an automated structure-based docking pipeline built around a set of tools developed in our team. This exercise again demonstrated a crucial importance of the correct local ligand geometry for the overall success of docking. Starting from the second part of the pose prediction stage, we developed a stable pipeline for sampling macrocycle conformers. This resulted in the subangstrom average precision of our pose predictions. In the affinity prediction exercise we obtained average results. However, we could improve these when using docking poses submitted by the best predictors. Our docking tools including the Convex-PL scoring function are available at https://team.inria.fr/nano-d/software/.

4D- quantitative structure-activity relationship modeling: making a comeback

Introduction: Predictive Quantitative Structure-Activity Relationship (QSAR) modeling has become an essential methodology for rapidly assessing various properties of chemicals. The vast majority of these QSAR models utilize numerical descriptors derived from the two- and/or three-dimensional structures of molecules. However, the conformation-dependent characteristics of flexible molecules and their dynamic interactions with biological target(s) is/are not encoded by these descriptors, leading to limited prediction performances and reduced interpretability. 2D/3D QSAR models are successful for virtual screening, but typically suffer at lead optimization stages. That is why conformation-dependent 4D-QSAR modeling methods were developed two decades ago. However, these methods have always suffered from the associated computational cost. Recently, 4D-QSAR has been experiencing a significant come-back due to rapid advances in GPU-accelerated molecular dynamic simulations and modern machine learning techniques. Areas covered: Herein, the authors briefly review the literature regarding 4D-QSAR modeling and describe its modern workflow called MD-QSAR. Challenges and current limitations are also highlighted. Expert opinion: The development of hyper-predictive MD-QSAR models could represent a disruptive technology for analyzing, understanding, and optimizing dynamic protein-ligand interactions with countless applications for drug discovery and chemical toxicity assessment. Therefore, there has never been a better time and relevance for molecular modeling teams to engage in hyper-predictive MD-QSAR modeling.

Computer-Assisted Selective Optimization of Side-Activities-from Cinalukast to a PPARα Modulator

Automated computational analogue design and scoring can speed up hit-to-lead optimization and appears particularly promising in selective optimization of side-activities (SOSA) where possible analogue diversity is confined. Probing this concept, we employed the cysteinyl leukotriene receptor 1 (CysLT₁ R) antagonist cinalukast as lead for which we discovered peroxisome proliferator-activated receptor α (PPARα) modulatory activity. We automatically generated a virtual library of close analogues and classified these roughly 8000 compounds for PPARα agonism and CysLT₁ R antagonism using automated affinity scoring and machine learning. A computationally preferred analogue for SOSA was synthesized, and in vitro characterization indeed revealed a marked activity shift toward enhanced PPARα activation and diminished CysLT₁ R antagonism. Thereby, this prospective application study highlights the potential of automating SOSA.

Continuous Evaluation of Ligand Protein Predictions: A Weekly Community Challenge for Drug Docking

Docking calculations can accelerate drug discovery by predicting the bound poses of ligands for a targeted protein. However, it is not clear which docking methods work best. Furthermore, predicting poses requires steps outside the docking algorithm itself, such as preparation of the protein and ligand, and it is not known which components are most in need of improvement. The Continuous Evaluation of Ligand Protein Predictions (CELPP) is a blinded prediction challenge designed to address these issues. Participants create a workflow to predict protein-ligand binding poses, which is then tasked with predicting 10-100 new protein-ligand crystal structures each week. CELPP evaluates the accuracy of each workflow's predictions and posts the scores online. The results can be used to identify the strengths and weaknesses of current approaches, help map docking problems to the algorithms most likely to overcome them, and illuminate areas of unmet need in structure-guided drug design.

Comprehensive assessment of flexible-ligand docking algorithms: current effectiveness and challenges

Protein-ligand docking has been playing an important role in modern drug discovery. To model drug-target binding in real systems, a number of flexible-ligand docking algorithms with different sampling strategies and scoring methods have been subsequently developed over the past three decades, while rigid-ligand docking is still being used because of its compelling computational efficiency. Here, a comprehensive assessment has been conducted to investigate the effectiveness of flexible-ligand docking versus rigid-ligand docking for three representative docking algorithms (global optimization, incremental construction and multi-conformer docking) in virtual screening and pose prediction on the Directory of Useful Decoys. It was found that overall flexible-ligand docking did not achieve a statistically significant improvement in enrichments over rigid-ligand docking in virtual screening, but all docking programs significantly improved the success rates when considering ligand flexibility in pose prediction. The worse effectiveness of flexible-ligand docking in virtual screening than in pose prediction suggests that the challenges of current docking algorithms exist in ranking more than docking, although the use of flexible-ligand docking in virtual screening was justified by its better effectiveness for more flexible ligand in virtual screening. Challenges for scoring, including internal energy, charge polarization, entropy and flexibility, were investigated and discussed. An empirical way was also proposed to consider loss of ligand conformational entropy for virtual screening.

Binding mode information improves fragment docking

Docking is commonly used in drug discovery to predict how ligand binds to protein target. Best programs are generally able to generate a correct solution, yet often fail to identify it. In the case of drug-like molecules, the correct and incorrect poses can be sorted by similarity to the crystallographic structure of the protein in complex with reference ligands. Fragments are particularly sensitive to scoring problems because they are weak ligands which form few interactions with protein. In the present study, we assessed the utility of binding mode information in fragment pose prediction. We compared three approaches: interaction fingerprints, 3D-matching of interaction patterns and 3D-matching of shapes. We prepared a test set composed of high-quality structures of the Protein Data Bank. We generated and evaluated the docking poses of 586 fragment/protein complexes. We observed that the best approach is twice as accurate as the native scoring function, and that post-processing is less effective for smaller fragments. Interestingly, fragments and drug-like molecules both proved to be useful references. In the discussion, we suggest the best conditions for a successful pose prediction with the three approaches.

In Need of Bias Control: Evaluating Chemical Data for Machine Learning in Structure-Based Virtual Screening

Reports of successful applications of machine learning (ML) methods in structure-based virtual screening (SBVS) are increasing. ML methods such as convolutional neural networks show promising results and often outperform traditional methods such as empirical scoring functions in retrospective validation. However, trained ML models are often treated as black boxes and are not straightforwardly interpretable. In most cases, it is unknown which features in the data are decisive and whether a model's predictions are right for the right reason. Hence, we re-evaluated three widely used benchmark data sets in the context of ML methods and came to the conclusion that not every benchmark data set is suitable. Moreover, we demonstrate on two examples from current literature that bias is learned implicitly and unnoticed from standard benchmarks. On the basis of these results, we conclude that there is a need for eligible validation experiments and benchmark data sets suited to ML for more bias-controlled validation in ML-based SBVS. Therefore, we provide guidelines for setting up validation experiments and give a perspective on how new data sets could be generated.

Predicting protein-ligand binding modes for CELPP and GC3: workflows and insight

Drug Design Data Resource (D3R) continues to release valuable benchmarking datasets to promote improvement and development of computational methods for new drug discovery. We have developed several methods for protein-ligand binding mode prediction during the participation in the D3R challenges. In the present study, these methods were integrated, automated, and systematically tested using the large-scale data from Continuous Evaluation of Ligand Pose Prediction (CELPP) and a subset of Grand challenge 3 (GC3). The results show that current molecular docking methods benefit from the increasing number of protein-ligand complex structures deposited in Protein Data Bank. Using an appropriate protein structure for docking significantly improves the success rate of the binding mode prediction. The results of our template-based method and docking method are compared and discussed. Our future direction include the combination of these two methods for binding mode prediction.

D3R Grand Challenge 3: blind prediction of protein-ligand poses and affinity rankings

The Drug Design Data Resource aims to test and advance the state of the art in protein-ligand modeling by holding community-wide blinded, prediction challenges. Here, we report on our third major round, Grand Challenge 3 (GC3). Held 2017-2018, GC3 centered on the protein Cathepsin S and the kinases VEGFR2, JAK2, p38-α, TIE2, and ABL1, and included both pose-prediction and affinity-ranking components. GC3 was structured much like the prior challenges GC2015 and GC2. First, Stage 1 tested pose prediction and affinity ranking methods; then all available crystal structures were released, and Stage 2 tested only affinity rankings, now in the context of the available structures. Unique to GC3 was the addition of a Stage 1b self-docking subchallenge, in which the protein coordinates from all of the cocrystal structures used in the cross-docking challenge were released, and participants were asked to predict the pose of CatS ligands using these newly released structures. We provide an overview of the outcomes and discuss insights into trends and best-practices.

Pushing the Limits of Computational Structure-Based Drug Design with a Cryo-EM Structure: The Ca2+ Channel α2δ-1 Subunit as a Test Case

Cryo-electron microscopy (cryo-EM) is emerging as a real alternative for structural elucidation. In spite of this, very few cryo-EM structures have been described so far as successful platforms for in silico drug design. Gabapentin and pregabalin are some of the most successful drugs in the treatment of epilepsy and neuropathic pain. Although both are in clinical use and are known to exert their effects by binding to the regulatory α2δ subunit of voltage gated calcium channels, their binding modes have never been characterized. We describe here the successful use of an exhaustive protein-ligand sampling algorithm on the α2δ-1 subunit of the recently published cryo-EM structure, with the goal of characterizing the ligand entry path and binding mode for gabapentin, pregabalin, and several other amino acidic α2δ-1 ligands. Our studies indicate that (i) all simulated drugs explore the same path for accessing the occluded binding site on the interior of the α2δ-1 subunit; (ii) they all roughly occupy the same pocket; (iii) the plasticity of the binding site allows the accommodation of a variety of amino acidic modulators, driven by the flexible "capping loop" delineated by residues Tyr426-Val435 and the floppy nature of Arg217; (iv) the predicted binding modes are in line with previously available mutagenesis data, confirming Arg217 as key for binding, with Asp428 and Asp467 highlighted as additional anchoring points for all amino acidic drugs. The study is one of the first proofs that latest-generation cryo-EM structures combined with exhaustive computational methods can be exploited in early drug discovery.

Convolutional neural network scoring and minimization in the D3R 2017 community challenge

We assess the ability of our convolutional neural network (CNN)-based scoring functions to perform several common tasks in the domain of drug discovery. These include correctly identifying ligand poses near and far from the true binding mode when given a set of reference receptors and classifying ligands as active or inactive using structural information. We use the CNN to re-score or refine poses generated using a conventional scoring function, Autodock Vina, and compare the performance of each of these methods to using the conventional scoring function alone. Furthermore, we assess several ways of choosing appropriate reference receptors in the context of the D3R 2017 community benchmarking challenge. We find that our CNN scoring function outperforms Vina on most tasks without requiring manual inspection by a knowledgeable operator, but that the pose prediction target chosen for the challenge, Cathepsin S, was particularly challenging for de novo docking. However, the CNN provided best-in-class performance on several virtual screening tasks, underscoring the relevance of deep learning to the field of drug discovery.

Boosted neural networks scoring functions for accurate ligand docking and ranking

Predicting the native poses of ligands correctly is one of the most important steps towards successful structure-based drug design. Binding affinities (BAs) estimated by traditional scoring functions (SFs) are typically used to score and rank-order poses to select the most promising conformation. This BA-based approach is widely applied and some success has been reported, but it is inconsistent and still far from perfect. The main reason for this is that SFs are trained on experimental BA values of only native poses found in co-crystallized structures of protein-ligand complexes (PLCs). However, during docking, they are needed to discriminate between native and decoy poses, a task for which they have not been specifically designed. To overcome this limitation, we propose to build task-specific SFs that model binding affinities (scoring task) as well as conformations (docking task) using the root mean square deviation (RMSD) of a ligand pose from the native pose. Our models are based on boosted ensembles of neural networks and other state-of-the-art machine learning (ML) algorithms in conjunction with multi-perspective interaction modeling techniques for accurate characterization of PLCs.

We assess the docking and scoring/ranking accuracies of the proposed ML SFs as well as three conventional SFs in the context of the 2014 CSAR benchmark exercise that encompasses three high-quality protein systems and a diverse set of drug-like molecules. Our proposed docking-specific SFs provide a substantial improvement in the docking task. We find that RMSD-based SFs for BsN, an ensemble neural networks (NN) model based on boosting, and six other ML models provide more than 120% improvement, on average, over their BA-based counterparts. In terms of scoring/ranking accuracy, we find that the approach of using RMSD-based BsN to select the top ligand pose followed by applying BA-based BsN to rank ligands using predicted BA scores leads to consistent and correctly ranked ligands for the two protein targets Spleen Tyrosine Kinase (SYK) and tRNA (m1G37) methyltransferase (TrmD). In addition, the ensemble NN SF BsN is at least 250% more accurate than a single neural network (SNN) model. We further find that ensemble models based on NNs surpass SFs based on other state-of-the-art ML algorithms such as BRT, RF, SVM, and [Formula: see text]NN. Finally, our RF model fitted to PLCs characterized by multiple sets of descriptors from four different sources (X-Score, AffiScore, RF-Score, and GOLD) substantially outperforms the SF RF-Score that uses only one set of features, underlining the value of multi-perspective modeling.

Plasticity of the Binding Site of Renin: Optimized Selection of Protein Structures for Ensemble Docking

Protein flexibility poses a major challenge to docking of potential ligands in that the binding site can adopt different shapes. Docking algorithms usually keep the protein rigid and only allow the ligand to be treated as flexible. However, a wrong assessment of the shape of the binding pocket can prevent a ligand from adapting a correct pose. Ensemble docking is a simple yet promising method to solve this problem: Ligands are docked into multiple structures, and the results are subsequently merged. Selection of protein structures is a significant factor for this approach. In this work we perform a comprehensive and comparative study evaluating the impact of structure selection on ensemble docking. We perform ensemble docking with several crystal structures and with structures derived from molecular dynamics simulations of renin, an attractive target for antihypertensive drugs. Here, 500 ns of MD simulations revealed binding site shapes not found in any available crystal structure. We evaluate the importance of structure selection for ensemble docking by comparing binding pose prediction, ability to rank actives above nonactives (screening utility), and scoring accuracy. As a result, for ensemble definition k-means clustering appears to be better suited than hierarchical clustering with average linkage. The best performing ensemble consists of four crystal structures and is able to reproduce the native ligand poses better than any individual crystal structure. Moreover this ensemble outperforms 88% of all individual crystal structures in terms of screening utility as well as scoring accuracy. Similarly, ensembles of MD-derived structures perform on average better than 75% of any individual crystal structure in terms of scoring accuracy at all inspected ensembles sizes.

RosettaLigandEnsemble: A Small-Molecule Ensemble-Driven Docking Approach

RosettaLigand is a protein-small-molecule (ligand) docking software capable of predicting binding poses and is used for virtual screening of medium-sized ligand libraries. Structurally similar small molecules are generally found to bind in the same pose to one binding pocket, despite some prominent exceptions. To make use of this information, we have developed RosettaLigandEnsemble (RLE). RLE docks a superimposed ensemble of congeneric ligands simultaneously. The program determines a well-scoring overall pose for this superimposed ensemble before independently optimizing individual protein-small-molecule interfaces. In a cross-docking benchmark of 89 protein-small-molecule co-crystal structures across 20 biological systems, we found that RLE improved sampling efficiency in 62 cases, with an average change of 18%. In addition, RLE generated more consistent docking results within a congeneric series and was capable of rescuing the unsuccessful docking of individual ligands, identifying a nativelike top-scoring model in 10 additional cases. The improvement in RLE is driven by a balance between having a sizable common chemical scaffold and meaningful modifications to distal groups. The new ensemble docking algorithm will work well in conjunction with medicinal chemistry structure-activity relationship (SAR) studies to more accurately recapitulate protein-ligand interfaces. We also tested whether optimizing the rank correlation of RLE-binding scores to SAR data in the refinement step helps the high-resolution positioning of the ligand. However, no significant improvement was observed.