CSAR data set release 2012: ligands, affinities, complexes, and docking decoys

Addressing docking pose selection with structure-based deep learning: Recent advances, challenges and opportunities

Molecular docking is a widely used technique in drug discovery to predict the binding mode of a given ligand to its target. However, the identification of the near-native binding pose in docking experiments still represents a challenging task as the scoring functions currently employed by docking programs are parametrized to predict the binding affinity, and, therefore, they often fail to correctly identify the ligand native binding conformation. Selecting the correct binding mode is crucial to obtaining meaningful results and to conveniently optimizing new hit compounds. Deep learning (DL) algorithms have been an area of a growing interest in this sense for their capability to extract the relevant information directly from the protein-ligand structure. Our review aims to present the recent advances regarding the development of DL-based pose selection approaches, discussing limitations and possible future directions. Moreover, a comparison between the performances of some classical scoring functions and DL-based methods concerning their ability to select the correct binding mode is reported. In this regard, two novel DL-based pose selectors developed by us are presented.

Binding interactions and in silico ADME prediction of isoconessimine derivatives as potent acetylcholinesterase inhibitors

In pursuit of new acetylcholinesterase (AChE) inhibitors for treating Alzheimer's disease (AD), a series of ten previously synthesized isoconessimine compounds (7a-7j) was in silico investigated for their binding interactions with AChE and pharmacokinetics based on absorption, distribution, metabolism, and excretion (ADME) properties using molecular docking, ONIOM (Our own N-layered Integrated molecular Orbital and molecular Mechanics) method and SwissADME tools. Docking experiments showed that all compounds bind within the active site gorge of AChE (PDB entry 1C2B), posing its aryloxy-substitutional ethyl group to catalytic site and conessine skeleton to peripheral anionic site. ONIOM interaction energy was used as an ONIOM score to improve docking score, and it ranked 7b as the most potent AChE inhibitor, in agreement with previous experiment. Residues, ASP74, TRP86, GLY122, GLU202, TRP286, GLU292, SER293, ILE294, TYR337, TYR341, and HIS447 were identified as important for the binding of the AChE-isoconessimine complex. The SwissADME investigation suggested that four compounds (7a, 7c, 7d and 7f) agree with the rules of drug-likeness. The steric and electronic effects on the aryloxy-substitutional ethyl group as important factors in the AChE inhibition were also discussed, which brings a better understanding of Alzheimer's disease drug development.

Structure-based, deep-learning models for protein-ligand binding affinity prediction

The launch of AlphaFold series has brought deep-learning techniques into the molecular structural science. As another crucial problem, structure-based prediction of protein-ligand binding affinity urgently calls for advanced computational techniques. Is deep learning ready to decode this problem? Here we review mainstream structure-based, deep-learning approaches for this problem, focusing on molecular representations, learning architectures and model interpretability. A model taxonomy has been generated. To compensate for the lack of valid comparisons among those models, we realized and evaluated representatives from a uniform basis, with the advantages and shortcomings discussed. This review will potentially benefit structure-based drug discovery and related areas.

Multi-task bioassay pre-training for protein-ligand binding affinity prediction

Protein-ligand binding affinity (PLBA) prediction is the fundamental task in drug discovery. Recently, various deep learning-based models predict binding affinity by incorporating the three-dimensional (3D) structure of protein-ligand complexes as input and achieving astounding progress. However, due to the scarcity of high-quality training data, the generalization ability of current models is still limited. Although there is a vast amount of affinity data available in large-scale databases such as ChEMBL, issues such as inconsistent affinity measurement labels (i.e. IC50, Ki, Kd), different experimental conditions, and the lack of available 3D binding structures complicate the development of high-precision affinity prediction models using these data. To address these issues, we (i) propose Multi-task Bioassay Pre-training (MBP), a pre-training framework for structure-based PLBA prediction; (ii) construct a pre-training dataset called ChEMBL-Dock with more than 300k experimentally measured affinity labels and about 2.8M docked 3D structures. By introducing multi-task pre-training to treat the prediction of different affinity labels as different tasks and classifying relative rankings between samples from the same bioassay, MBP learns robust and transferrable structural knowledge from our new ChEMBL-Dock dataset with varied and noisy labels. Experiments substantiate the capability of MBP on the structure-based PLBA prediction task. To the best of our knowledge, MBP is the first affinity pre-training model and shows great potential for future development. MBP web-server is now available for free at: https://huggingface.co/spaces/jiaxianustc/mbp.

The Impact of Data on Structure-Based Binding Affinity Predictions Using Deep Neural Networks

Artificial intelligence (AI) has gained significant traction in the field of drug discovery, with deep learning (DL) algorithms playing a crucial role in predicting protein-ligand binding affinities. Despite advancements in neural network architectures, system representation, and training techniques, the performance of DL affinity prediction has reached a plateau, prompting the question of whether it is truly solved or if the current performance is overly optimistic and reliant on biased, easily predictable data. Like other DL-related problems, this issue seems to stem from the training and test sets used when building the models. In this work, we investigate the impact of several parameters related to the input data on the performance of neural network affinity prediction models. Notably, we identify the size of the binding pocket as a critical factor influencing the performance of our statistical models; furthermore, it is more important to train a model with as much data as possible than to restrict the training to only high-quality datasets. Finally, we also confirm the bias in the typically used current test sets. Therefore, several types of evaluation and benchmarking are required to understand models' decision-making processes and accurately compare the performance of models.

PLAS-5k: Dataset of Protein-Ligand Affinities from Molecular Dynamics for Machine Learning Applications

Computational methods and recently modern machine learning methods have played a key role in structure-based drug design. Though several benchmarking datasets are available for machine learning applications in virtual screening, accurate prediction of binding affinity for a protein-ligand complex remains a major challenge. New datasets that allow for the development of models for predicting binding affinities better than the state-of-the-art scoring functions are important. For the first time, we have developed a dataset, PLAS-5k comprised of 5000 protein-ligand complexes chosen from PDB database. The dataset consists of binding affinities along with energy components like electrostatic, van der Waals, polar and non-polar solvation energy calculated from molecular dynamics simulations using MMPBSA (Molecular Mechanics Poisson-Boltzmann Surface Area) method. The calculated binding affinities outperformed docking scores and showed a good correlation with the available experimental values. The availability of energy components may enable optimization of desired components during machine learning-based drug design. Further, OnionNet model has been retrained on PLAS-5k dataset and is provided as a baseline for the prediction of binding affinities.

Transition Rates and Efficiency of Collective Variables from Time-Dependent Biased Simulations

Simulations with adaptive time-dependent bias enable an efficient exploration of the conformational space of a system. However, the dynamic information is altered by the bias. Infrequent metadynamics recovers the transition rate of crossing a barrier, if the collective variables are ideal and there is no bias deposition near the transition state. Unfortunately, these conditions are not always fulfilled. To overcome these limitations, and inspired by single-molecule force spectroscopy, we use Kramers' theory for calculating the barrier-crossing rate when a time-dependent bias is added to the system. We assess the efficiency of collective variables parameter by measuring how efficiently the bias accelerates the transitions. We present approximate analytical expressions of the survival probability, reproducing the barrier-crossing time statistics and enabling the extraction of the unbiased transition rate even for challenging cases. We explore the limits of our method and provide convergence criteria to assess its validity.

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.

Sfcnn: a novel scoring function based on 3D convolutional neural network for accurate and stable protein-ligand affinity prediction

Background

Computer-aided drug design provides an effective method of identifying lead compounds. However, success rates are significantly bottlenecked by the lack of accurate and reliable scoring functions needed to evaluate binding affinities of protein–ligand complexes. Therefore, many scoring functions based on machine learning or deep learning have been developed to improve prediction accuracies in recent years. In this work, we proposed a novel featurization method, generating a new scoring function model based on 3D convolutional neural network.

Results

This work showed the results from testing four architectures and three featurization methods, and outlined the development of a novel deep 3D convolutional neural network scoring function model. This model simplified feature engineering, and in combination with Grad-CAM made the intermediate layers of the neural network more interpretable. This model was evaluated and compared with other scoring functions on multiple independent datasets. The Pearson correlation coefficients between the predicted binding affinities by our model and the experimental data achieved 0.7928, 0.7946, 0.6758, and 0.6474 on CASF-2016 dataset, CASF-2013 dataset, CSAR_HiQ_NRC_set, and Astex_diverse_set, respectively. Overall, our model performed accurately and stably enough in the scoring power to predict the binding affinity of a protein–ligand complex.

Conclusions

These results indicate our model is an excellent scoring function, and performs well in scoring power for accurately and stably predicting the protein–ligand affinity. Our model will contribute towards improving the success rate of virtual screening, thus will accelerate the development of potential drugs or novel biologically active lead compounds.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12859-022-04762-3.

Artificial intelligence in virtual screening: models versus experiments

A typical drug discovery project involves identifying active compounds with significant binding potential for selected disease-specific targets. Experimental high-throughput screening (HTS) is a traditional approach to drug discovery, but is expensive and time-consuming when dealing with huge chemical libraries with billions of compounds. The search space can be narrowed down with the use of reliable computational screening approaches. In this review, we focus on various machine-learning (ML) and deep-learning (DL)-based scoring functions developed for solving classification and ranking problems in drug discovery. We highlight studies in which ML and DL models were successfully deployed to identify lead compounds for which the experimental validations are available from bioassay studies.

Combined Free-Energy Calculation and Machine Learning Methods for Understanding Ligand Unbinding Kinetics

The determination of drug residence times, which define the time an inhibitor is in complex with its target, is a fundamental part of the drug discovery process. Synthesis and experimental measurements of kinetic rate constants are, however, expensive and time consuming. In this work, we aimed to obtain drug residence times computationally. Furthermore, we propose a novel algorithm to identify molecular design objectives based on ligand unbinding kinetics. We designed an enhanced sampling technique to accurately predict the free-energy profiles of the ligand unbinding process, focusing on the free-energy barrier for unbinding. Our method first identifies unbinding paths determining a corresponding set of internal coordinates (ICs) that form contacts between the protein and the ligand; it then iteratively updates these interactions during a series of biased molecular dynamics (MD) simulations to reveal the ICs that are important for the whole of the unbinding process. Subsequently, we performed finite-temperature string simulations to obtain the free-energy barrier for unbinding using the set of ICs as a complex reaction coordinate. Importantly, we also aimed to enable the further design of drugs focusing on improved residence times. To this end, we developed a supervised machine learning (ML) approach with inputs from unbiased “downhill” trajectories initiated near the transition state (TS) ensemble of the string unbinding path. We demonstrate that our ML method can identify key ligand–protein interactions driving the system through the TS. Some of the most important drugs for cancer treatment are kinase inhibitors. One of these kinase targets is cyclin-dependent kinase 2 (CDK2), an appealing target for anticancer drug development. Here, we tested our method using two different CDK2 inhibitors for the potential further development of these compounds. We compared the free-energy barriers obtained from our calculations with those observed in available experimental data. We highlighted important interactions at the distal ends of the ligands that can be targeted for improved residence times. Our method provides a new tool to determine unbinding rates and to identify key structural features of the inhibitors that can be used as starting points for novel design strategies in drug discovery.

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

One of the main challenges of structure-based virtual screening (SBVS) is the incorporation of the receptor's flexibility, as its explicit representation in every docking run implies a high computational cost. Therefore, a common alternative to include the receptor's flexibility is the approach known as ensemble docking. Ensemble docking consists of using a set of receptor conformations and performing the docking assays over each of them. However, there is still no agreement on how to combine the ensemble docking results to obtain the final ligand ranking. A common choice is to use consensus strategies to aggregate the ensemble docking scores, but these strategies exhibit slight improvement regarding the single-structure approach. Here, we claim that using machine learning (ML) methodologies over the ensemble docking results could improve the predictive power of SBVS. To test this hypothesis, four proteins were selected as study cases: CDK2, FXa, EGFR, and HSP90. Protein conformational ensembles were built from crystallographic structures, whereas the evaluated compound library comprised up to three benchmarking data sets (DUD, DEKOIS 2.0, and CSAR-2012) and cocrystallized molecules. Ensemble docking results were processed through 30 repetitions of 4-fold cross-validation to train and validate two ML classifiers: logistic regression and gradient boosting trees. Our results indicate that the ML classifiers significantly outperform traditional consensus strategies and even the best performance case achieved with single-structure docking. We provide statistical evidence that supports the effectiveness of ML to improve the ensemble docking performance.

Binding affinity prediction for protein-ligand complex using deep attention mechanism based on intermolecular interactions

BACKGROUND

Accurate prediction of protein-ligand binding affinity is important for lowering the overall cost of drug discovery in structure-based drug design. For accurate predictions, many classical scoring functions and machine learning-based methods have been developed. However, these techniques tend to have limitations, mainly resulting from a lack of sufficient energy terms to describe the complex interactions between proteins and ligands. Recent deep-learning techniques can potentially solve this problem. However, the search for more efficient and appropriate deep-learning architectures and methods to represent protein-ligand complex is ongoing.

RESULTS

In this study, we proposed a deep-neural network model to improve the prediction accuracy of protein-ligand complex binding affinity. The proposed model has two important features, descriptor embeddings with information on the local structures of a protein-ligand complex and an attention mechanism to highlight important descriptors for binding affinity prediction. The proposed model performed better than existing binding affinity prediction models on most benchmark datasets.

CONCLUSIONS

We confirmed that an attention mechanism can capture the binding sites in a protein-ligand complex to improve prediction performance. Our code is available at https://github.com/Blue1993/BAPA .

Water molecules at protein-drug interfaces: computational prediction and analysis methods

The fundamental importance of water molecules at drug-protein interfaces is now widely recognised and a significant feature in structure-based drug design. Experimental methods for analysing the role of water in drug binding have many challenges, including the accurate location of bound water molecules in crystal structures, and problems in resolving specific water contributions to binding thermodynamics. Computational analyses of binding site water molecules provide an alternative, and in principle complete, structural and thermodynamic picture, and their use is now commonplace in the pharmaceutical industry. In this review, we describe the computational methodologies that are available and discuss their strengths and weaknesses. Additionally, we provide a critical analysis of the experimental data used to validate the methods, regarding the type and quality of experimental structural data. We also discuss some of the fundamental difficulties of each method and suggest directions for future study.

The critical role of QM/MM X-ray refinement and accurate tautomer/protomer determination in structure-based drug design

Conventional protein:ligand crystallographic refinement uses stereochemistry restraints coupled with a rudimentary energy functional to ensure the correct geometry of the model of the macromolecule-along with any bound ligand(s)-within the context of the experimental, X-ray density. These methods generally lack explicit terms for electrostatics, polarization, dispersion, hydrogen bonds, and other key interactions, and instead they use pre-determined parameters (e.g. bond lengths, angles, and torsions) to drive structural refinement. In order to address this deficiency and obtain a more complete and ultimately more accurate structure, we have developed an automated approach for macromolecular refinement based on a two layer, QM/MM (ONIOM) scheme as implemented within our DivCon Discovery Suite and "plugged in" to two mainstream crystallographic packages: PHENIX and BUSTER. This implementation is able to use one or more region layer(s), which is(are) characterized using linear-scaling, semi-empirical quantum mechanics, followed by a system layer which includes the balance of the model and which is described using a molecular mechanics functional. In this work, we applied our Phenix/DivCon refinement method-coupled with our XModeScore method for experimental tautomer/protomer state determination-to the characterization of structure sets relevant to structure-based drug design (SBDD). We then use these newly refined structures to show the impact of QM/MM X-ray refined structure on our understanding of function by exploring the influence of these improved structures on protein:ligand binding affinity prediction (and we likewise show how we use post-refinement scoring outliers to inform subsequent X-ray crystallographic efforts). Through this endeavor, we demonstrate a computational chemistry ↔ structural biology (X-ray crystallography) "feedback loop" which has utility in industrial and academic pharmaceutical research as well as other allied fields.

Dynamic Docking Using Multicanonical Molecular Dynamics: Simulating Complex Formation at the Atomistic Level

Multicanonical molecular dynamics (McMD)-based dynamic docking has been applied to predict the native binding configurations for several protein receptors and their ligands. Due to the enhanced sampling capabilities of McMD, it can exhaustively sample bound and unbound ligand configurations, as well as receptor conformations, and thus enables efficient sampling of the conformational and configurational space, not possible using canonical MD simulations. As McMD samples a wide configurational space, extensive analysis is required to study the diverse ensemble consisting of bound and unbound structures. By projecting the reweighted ensemble onto the first two principal axes obtained via principal component analysis of the multicanonical ensemble, the free energy landscape (FEL) can be obtained. Further analysis produces representative structures positioned at the local minima of the FEL, where these structures are then ranked by their free energy. In this chapter, we describe our dynamic docking methodology, which has successfully reproduced the native binding configuration for small compounds, medium-sized compounds, and peptide molecules.

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.

Combining Monte Carlo and Molecular Dynamics Simulations for Enhanced Binding Free Energy Estimation through Markov State Models

We present a multistep protocol, combining Monte Carlo and molecular dynamics simulations, for the estimation of absolute binding free energies, one of the most significant challenges in computer-aided drug design. The protocol is based on an initial short enhanced Monte Carlo simulation, followed by clustering of the ligand positions, which serve to identify the most relevant states of the unbinding process. From these states, extensive molecular dynamics simulations are run to estimate an equilibrium probability distribution obtained with Markov State Models, which is subsequently used to estimate the binding free energy. We tested the procedure on two different protein systems, the Plasminogen kringle domain 1 and Urokinase, each with multiple ligands, for an aggregated molecular dynamics length of 760 μs. Our results indicate that the initial sampling of the unbinding events largely facilitates the convergence of the subsequent molecular dynamics exploration. Moreover, the protocol is capable to properly rank the set of ligands examined, albeit with a significant computational cost for the, more realistic, Urokinase complexes. Overall, this work demonstrates the usefulness of combining enhanced sampling methods with regular simulation techniques as a way to obtain more reliable binding affinity estimates.

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.

Machine learning and ligand binding predictions: A review of data, methods, and obstacles

Computational predictions of ligand binding is a difficult problem, with more accurate methods being extremely computationally expensive. The use of machine learning for drug binding predictions could possibly leverage the use of biomedical big data in exchange for time-intensive simulations. This paper reviews current trends in the use of machine learning for drug binding predictions, data sources to develop machine learning algorithms, and potential problems that may lead to overfitting and ungeneralizable models. A few popular datasets that can be used to develop virtual high-throughput screening models are characterized using spatial statistics to quantify potential biases. We can see from evaluating some common benchmarks that good performance correlates with models with high-predicted bias scores and models with low bias scores do not have much predictive power. A better understanding of the limits of available data sources and how to fix them will lead to more generalizable models that will lead to novel drug discovery.

Cross-docking benchmark for automated pose and ranking prediction of ligand binding

Significant efforts have been devoted in the last decade to improving molecular docking techniques to predict both accurate binding poses and ranking affinities. Some shortcomings in the field are the limited number of standard methods for measuring docking success and the availability of widely accepted standard data sets for use as benchmarks in comparing different docking algorithms throughout the field. In order to address these issues, we have created a Cross-Docking Benchmark server. The server is a versatile cross-docking data set containing 4,399 protein-ligand complexes across 95 protein targets intended to serve as benchmark set and gold standard for state-of-the-art pose and ranking prediction in easy, medium, hard, or very hard docking targets. The benchmark along with a customizable cross-docking data set generation tool is available at http://disco.csb.pitt.edu. We further demonstrate the potential uses of the server in questions outside of basic benchmarking such as the selection of the ideal docking reference structure.

Comparison Study of Computational Prediction Tools for Drug-Target Binding Affinities

The drug development is generally arduous, costly, and success rates are low. Thus, the identification of drug-target interactions (DTIs) has become a crucial step in early stages of drug discovery. Consequently, developing computational approaches capable of identifying potential DTIs with minimum error rate are increasingly being pursued. These computational approaches aim to narrow down the search space for novel DTIs and shed light on drug functioning context. Most methods developed to date use binary classification to predict if the interaction between a drug and its target exists or not. However, it is more informative but also more challenging to predict the strength of the binding between a drug and its target. If that strength is not sufficiently strong, such DTI may not be useful. Therefore, the methods developed to predict drug-target binding affinities (DTBA) are of great value. In this study, we provide a comprehensive overview of the existing methods that predict DTBA. We focus on the methods developed using artificial intelligence (AI), machine learning (ML), and deep learning (DL) approaches, as well as related benchmark datasets and databases. Furthermore, guidance and recommendations are provided that cover the gaps and directions of the upcoming work in this research area. To the best of our knowledge, this is the first comprehensive comparison analysis of tools focused on DTBA with reference to AI/ML/DL.

Exploring fragment-based target-specific ranking protocol with machine learning on cathepsin S

Cathepsin S (CatS), a member of cysteine cathepsin proteases, has been well studied due to its significant role in many pathological processes, including arthritis, cancer and cardiovascular diseases. CatS inhibitors have been included in D3R-GC3 for both docking pose prediction and affinity ranking, and in D3R-GC4 for binding affinity ranking. The difficulties posed by CatS inhibitors in D3R mainly come from three aspects: large size, high flexibility and similar chemical structures. We have participated in GC4; our best submitted model, which employs a similarity-based alignment docking and Vina scoring protocol, yielded Kendall's τ of 0.23 for 459 binders in GC4. In our further explorations with machine learning, by curating a CatS specific training set, adopting a similarity-based constrained docking method as well as an arm-based fragmentation strategy which can describe large inhibitors in a locality-sensitive fashion, our best structure-based ranking protocol can achieve Kendall's τ of 0.52 for all binders in GC4. In this exploration process, we have demonstrated the importance of training data, docking approaches and fragmentation strategies in inhibitor-ranking protocol development with machine learning.

Key Topics in Molecular Docking for Drug Design

Molecular docking has been widely employed as a fast and inexpensive technique in the past decades, both in academic and industrial settings. Although this discipline has now had enough time to consolidate, many aspects remain challenging and there is still not a straightforward and accurate route to readily pinpoint true ligands among a set of molecules, nor to identify with precision the correct ligand conformation within the binding pocket of a given target molecule. Nevertheless, new approaches continue to be developed and the volume of published works grows at a rapid pace. In this review, we present an overview of the method and attempt to summarise recent developments regarding four main aspects of molecular docking approaches: (i) the available benchmarking sets, highlighting their advantages and caveats, (ii) the advances in consensus methods, (iii) recent algorithms and applications using fragment-based approaches, and (iv) the use of machine learning algorithms in molecular docking. These recent developments incrementally contribute to an increase in accuracy and are expected, given time, and together with advances in computing power and hardware capability, to eventually accomplish the full potential of this area.

DeepBindRG: a deep learning based method for estimating effective protein-ligand affinity

Proteins interact with small molecules to modulate several important cellular functions. Many acute diseases were cured by small molecule binding in the active site of protein either by inhibition or activation. Currently, there are several docking programs to estimate the binding position and the binding orientation of protein–ligand complex. Many scoring functions were developed to estimate the binding strength and predict the effective protein–ligand binding. While the accuracy of current scoring function is limited by several aspects, the solvent effect, entropy effect, and multibody effect are largely ignored in traditional machine learning methods. In this paper, we proposed a new deep neural network-based model named DeepBindRG to predict the binding affinity of protein–ligand complex, which learns all the effects, binding mode, and specificity implicitly by learning protein–ligand interface contact information from a large protein–ligand dataset. During the initial data processing step, the critical interface information was preserved to make sure the input is suitable for the proposed deep learning model. While validating our model on three independent datasets, DeepBindRG achieves root mean squared error (RMSE) value of pKa (−logK_d or −logK_i) about 1.6–1.8 and R value around 0.5–0.6, which is better than the autodock vina whose RMSE value is about 2.2–2.4 and R value is 0.42–0.57. We also explored the detailed reasons for the performance of DeepBindRG, especially for several failed cases by vina. Furthermore, DeepBindRG performed better for four challenging datasets from DUD.E database with no experimental protein–ligand complexes. The better performance of DeepBindRG than autodock vina in predicting protein–ligand binding affinity indicates that deep learning approach can greatly help with the drug discovery process. We also compare the performance of DeepBindRG with a 4D based deep learning method “pafnucy”, the advantage and limitation of both methods have provided clues for improving the deep learning based protein–ligand prediction model in the future.

Local Interaction Density (LID), a Fast and Efficient Tool to Prioritize Docking Poses

Ligand docking at a protein site can be improved by prioritizing poses by similarity to validated binding modes found in the crystal structures of ligand/protein complexes. The interactions formed in the predicted model are searched in each of the reference 3D structures, taken individually. We propose to merge the information provided by all references, creating a single representation of all known binding modes. The method is called LID, an acronym for Local Interaction Density. LID was benchmarked in a pose prediction exercise on 19 proteins and 1382 ligands using PLANTS as docking software. It was also tested in a virtual screening challenge on eight proteins, with a dataset of 140,000 compounds from DUD-E and PubChem. LID significantly improved the performance of the docking program in both pose prediction and virtual screening. The gain is comparable to that obtained with a rescoring approach based on the individual comparison of reference binding modes (the GRIM method). Importantly, LID is effective with a small number of references. LID calculation time is negligible compared to the docking time.

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.

Visualizing convolutional neural network protein-ligand scoring

Protein-ligand scoring is an important step in a structure-based drug design pipeline. Selecting a correct binding pose and predicting the binding affinity of a protein-ligand complex enables effective virtual screening. Machine learning techniques can make use of the increasing amounts of structural data that are becoming publicly available. Convolutional neural network (CNN) scoring functions in particular have shown promise in pose selection and affinity prediction for protein-ligand complexes. Neural networks are known for being difficult to interpret. Understanding the decisions of a particular network can help tune parameters and training data to maximize performance. Visualization of neural networks helps decompose complex scoring functions into pictures that are more easily parsed by humans. Here we present three methods for visualizing how individual protein-ligand complexes are interpreted by 3D convolutional neural networks. We also present a visualization of the convolutional filters and their weights. We describe how the intuition provided by these visualizations aids in network design.

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.

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.

Design of a heme-binding peptide motif adopting a β-hairpin conformation

Heme-binding proteins constitute a large family of catalytic and transport proteins. Their widespread presence as globins and as essential oxygen and electron transporters, along with their diverse enzymatic functions, have made them targets for protein design. Most previously reported designs involved the use of α-helical scaffolds, and natural peptides also exhibit a strong preference for these scaffolds. However, the reason for this preference is not well-understood, in part because alternative protein designs, such as those with β-sheets or hairpins, are challenging to perform. Here, we report the computational design and experimental validation of a water-soluble heme-binding peptide, Pincer-1, composed of predominantly β-scaffold secondary structures. Such heme-binding proteins are rarely observed in nature, and by designing such a scaffold, we simultaneously increase the known fold space of heme-binding proteins and expand the limits of computational design methods. For a β-scaffold, two tryptophan zipper β-hairpins sandwiching a heme molecule were linked through an N-terminal cysteine disulfide bond. β-Hairpin orientations and residue selection were performed computationally. Heme binding was confirmed through absorbance experiments and surface plasmon resonance experiments (K_D = 730 ± 160 nm). CD and NMR experiments validated the β-hairpin topology of the designed peptide. Our results indicate that a helical scaffold is not essential for heme binding and reveal the first designed water-soluble, heme-binding β-hairpin peptide. This peptide could help expand the search for and design space to cytoplasmic heme-binding proteins.

Estimating Electron Density Support for Individual Atoms and Molecular Fragments in X-ray Structures

Macromolecular structures resolved by X-ray crystallography are essential for life science research. While some methods exist to automatically quantify the quality of the electron density fit, none of them is without flaws. Especially the question of how well individual parts like atoms, small fragments, or molecules are supported by electron density is difficult to quantify. While taking experimental uncertainties correctly into account, they do not offer an answer on how reliable an individual atom position is. A rapid quantification of this atomic position reliability would be highly valuable in structure-based molecular design. To overcome this limitation, we introduce the electron density score EDIA for individual atoms and molecular fragments. EDIA assesses rapidly, automatically, and intuitively the fit of individual as well as multiple atoms (EDIA_m) into electron density accompanied by an integrated error analysis. The computation is based on the standard 2fo - fc electron density map in combination with the model of the molecular structure. For evaluating partial structures, EDIA_m shows significant advantages compared to the real-space R correlation coefficient (RSCC) and the real-space difference density Z score (RSZD) from the molecular modeler's point of view. Thus, EDIA abolishes the time-consuming step of visually inspecting the electron density during structure selection and curation. It supports daily modeling tasks of medicinal and computational chemists and enables a fully automated assembly of large-scale, high-quality structure data sets. Furthermore, EDIA scores can be applied for model validation and method development in computer-aided molecular design. In contrast to measuring the deviation from the structure model by root-mean-squared deviation, EDIA scores allow comparison to the underlying experimental data taking its uncertainty into account.

Ranking docking poses by graph matching of protein-ligand interactions: lessons learned from the D3R Grand Challenge 2

A novel docking challenge has been set by the Drug Design Data Resource (D3R) in order to predict the pose and affinity ranking of a set of Farnesoid X receptor (FXR) agonists, prior to the public release of their bound X-ray structures and potencies. In a first phase, 36 agonists were docked to 26 Protein Data Bank (PDB) structures of the FXR receptor, and next rescored using the in-house developed GRIM method. GRIM aligns protein-ligand interaction patterns of docked poses to those of available PDB templates for the target protein, and rescore poses by a graph matching method. In agreement with results obtained during the previous 2015 docking challenge, we clearly show that GRIM rescoring improves the overall quality of top-ranked poses by prioritizing interaction patterns already visited in the PDB. Importantly, this challenge enables us to refine the applicability domain of the method by better defining the conditions of its success. We notably show that rescoring apolar ligands in hydrophobic pockets leads to frequent GRIM failures. In the second phase, 102 FXR agonists were ranked by decreasing affinity according to the Gibbs free energy of the corresponding GRIM-selected poses, computed by the HYDE scoring function. Interestingly, this fast and simple rescoring scheme provided the third most accurate ranking method among 57 contributions. Although the obtained ranking is still unsuitable for hit to lead optimization, the GRIM-HYDE scoring scheme is accurate and fast enough to post-process virtual screening data.

Enhance the performance of current scoring functions with the aid of 3D protein-ligand interaction fingerprints

BACKGROUND

In structure-based drug design, binding affinity prediction remains as a challenging goal for current scoring functions. Development of target-biased scoring functions provides a new possibility for tackling this problem, but this approach is also associated with certain technical difficulties. We previously reported the Knowledge-Guided Scoring (KGS) method as an alternative approach (BMC Bioinformatics, 2010, 11, 193-208). The key idea is to compute the binding affinity of a given protein-ligand complex based on the known binding data of an appropriate reference complex, so the error in binding affinity prediction can be reduced effectively.

RESULTS

In this study, we have developed an upgraded version, i.e. KGS2, by employing 3D protein-ligand interaction fingerprints in reference selection. KGS2 was evaluated in combination with four scoring functions (X-Score, ChemPLP, ASP, and GoldScore) on five drug targets (HIV-1 protease, carbonic anhydrase 2, beta-secretase 1, beta-trypsin, and checkpoint kinase 1). In the in situ scoring test, considerable improvements were observed in most cases after application of KGS2. Besides, the performance of KGS2 was always better than KGS in all cases. In the more challenging molecular docking test, application of KGS2 also led to improved structure-activity relationship in some cases.

CONCLUSIONS

KGS2 can be applied as a convenient "add-on" to current scoring functions without the need to re-engineer them, and its application is not limited to certain target proteins as customized scoring functions. As an interpolation method, its accuracy in principle can be improved further with the increasing knowledge of protein-ligand complex structures and binding affinity data. We expect that KGS2 will become a practical tool for enhancing the performance of current scoring functions in binding affinity prediction. The KGS2 software is available upon contacting the authors.

Waterdock 2.0: Water placement prediction for Holo-structures with a pymol plugin

Water is often found to mediate interactions between a ligand and a protein. It can play a significant role in orientating the ligand within a binding pocket and contribute to the free energy of binding. It would thus be extremely useful to be able to accurately predict the position and orientation of water molecules within a binding pocket. Recently, we developed the WaterDock protocol that was able to predict 97% of the water molecules in a test set. However, this approach generated false positives at a rate of over 20% in most cases and whilst this might be acceptable for some applications, in high throughput scenarios this is not desirable. Here we tackle this problem via the inclusion of knowledge regarding the solvation structure of ligand functional groups. We call this new protocol WaterDock2 and demonstrate that this protocol maintains a similar true positive rate to the original implementation but is capable of reducing the false-positive rate by over 50%. To improve the usability of the method, we have also developed a plugin for the popular graphics program PyMOL. The plugin also contains an implementation of the original WaterDock.

Exploring the stability of ligand binding modes to proteins by molecular dynamics simulations

The binding mode prediction is of great importance to structure-based drug design. The discrimination of various binding poses of ligand generated by docking is a great challenge not only to docking score functions but also to the relatively expensive free energy calculation methods. Here we systematically analyzed the stability of various ligand poses under molecular dynamics (MD) simulation. First, a data set of 120 complexes was built based on the typical physicochemical properties of drug-like ligands. Three potential binding poses (one correct pose and two decoys) were selected for each ligand from self-docking in addition to the experimental pose. Then, five independent MD simulations for each pose were performed with different initial velocities for the statistical analysis. Finally, the stabilities of ligand poses under MD were evaluated and compared with the native one from crystal structure. We found that about 94% of the native poses were maintained stable during the simulations, which suggests that MD simulations are accurate enough to judge most experimental binding poses as stable properly. Interestingly, incorrect decoy poses were maintained much less and 38-44% of decoys could be excluded just by performing equilibrium MD simulations, though 56-62% of decoys were stable. The computationally-heavy binding free energy calculation can be performed only for these survived poses.

D3R grand challenge 2015: Evaluation of protein-ligand pose and affinity predictions

The Drug Design Data Resource (D3R) ran Grand Challenge 2015 between September 2015 and February 2016. Two targets served as the framework to test community docking and scoring methods: (1) HSP90, donated by AbbVie and the Community Structure Activity Resource (CSAR), and (2) MAP4K4, donated by Genentech. The challenges for both target datasets were conducted in two stages, with the first stage testing pose predictions and the capacity to rank compounds by affinity with minimal structural data; and the second stage testing methods for ranking compounds with knowledge of at least a subset of the ligand-protein poses. An additional sub-challenge provided small groups of chemically similar HSP90 compounds amenable to alchemical calculations of relative binding free energy. Unlike previous blinded Challenges, we did not provide cognate receptors or receptors prepared with hydrogens and likewise did not require a specified crystal structure to be used for pose or affinity prediction in Stage 1. Given the freedom to select from over 200 crystal structures of HSP90 in the PDB, participants employed workflows that tested not only core docking and scoring technologies, but also methods for addressing water-mediated ligand-protein interactions, binding pocket flexibility, and the optimal selection of protein structures for use in docking calculations. Nearly 40 participating groups submitted over 350 prediction sets for Grand Challenge 2015. This overview describes the datasets and the organization of the challenge components, summarizes the results across all submitted predictions, and considers broad conclusions that may be drawn from this collaborative community endeavor.