DeepPocket: Ligand Binding Site Detection and Segmentation using 3D Convolutional Neural Networks

LVPocket: integrated 3D global-local information to protein binding pockets prediction with transfer learning of protein structure classification

BACKGROUND

Previous deep learning methods for predicting protein binding pockets mainly employed 3D convolution, yet an abundance of convolution operations may lead the model to excessively prioritize local information, thus overlooking global information. Moreover, it is essential for us to account for the influence of diverse protein folding structural classes. Because proteins classified differently structurally exhibit varying biological functions, whereas those within the same structural class share similar functional attributes.

RESULTS

We proposed LVPocket, a novel method that synergistically captures both local and global information of protein structure through the integration of Transformer encoders, which help the model achieve better performance in binding pockets prediction. And then we tailored prediction models for data of four distinct structural classes of proteins using the transfer learning. The four fine-tuned models were trained on the baseline LVPocket model which was trained on the sc-PDB dataset. LVPocket exhibits superior performance on three independent datasets compared to current state-of-the-art methods. Additionally, the fine-tuned model outperforms the baseline model in terms of performance.

SCIENTIFIC CONTRIBUTION

We present a novel model structure for predicting protein binding pockets that provides a solution for relying on extensive convolutional computation while neglecting global information about protein structures. Furthermore, we tackle the impact of different protein folding structures on binding pocket prediction tasks through the application of transfer learning methods.

Highly accurate carbohydrate-binding site prediction with DeepGlycanSite

As the most abundant organic substances in nature, carbohydrates are essential for life. Understanding how carbohydrates regulate proteins in the physiological and pathological processes presents opportunities to address crucial biological problems and develop new therapeutics. However, the diversity and complexity of carbohydrates pose a challenge in experimentally identifying the sites where carbohydrates bind to and act on proteins. Here, we introduce a deep learning model, DeepGlycanSite, capable of accurately predicting carbohydrate-binding sites on a given protein structure. Incorporating geometric and evolutionary features of proteins into a deep equivariant graph neural network with the transformer architecture, DeepGlycanSite remarkably outperforms previous state-of-the-art methods and effectively predicts binding sites for diverse carbohydrates. Integrating with a mutagenesis study, DeepGlycanSite reveals the guanosine-5'-diphosphate-sugar-recognition site of an important G-protein coupled receptor. These findings demonstrate DeepGlycanSite is invaluable for carbohydrate-binding site prediction and could provide insights into molecular mechanisms underlying carbohydrate-regulation of therapeutically important proteins.

A comprehensive survey on protein-ligand binding site prediction

Protein-ligand binding site prediction is critical for protein function annotation and drug discovery. Biological experiments are time-consuming and require significant equipment, materials, and labor resources. Developing accurate and efficient computational methods for protein-ligand interaction prediction is essential. Here, we summarize the key challenges associated with ligand binding site (LBS) prediction and introduce recently published methods from their input features, computational algorithms, and ligand types. Furthermore, we investigate the specificity of allosteric site identification as a particular LBS type. Finally, we discuss the prospective directions for machine learning-based LBS prediction in the near future.

DrugMap: A quantitative pan-cancer analysis of cysteine ligandability

Cysteine-focused chemical proteomic platforms have accelerated the clinical development of covalent inhibitors for a wide range of targets in cancer. However, how different oncogenic contexts influence cysteine targeting remains unknown. To address this question, we have developed "DrugMap," an atlas of cysteine ligandability compiled across 416 cancer cell lines. We unexpectedly find that cysteine ligandability varies across cancer cell lines, and we attribute this to differences in cellular redox states, protein conformational changes, and genetic mutations. Leveraging these findings, we identify actionable cysteines in NF-κB1 and SOX10 and develop corresponding covalent ligands that block the activity of these transcription factors. We demonstrate that the NF-κB1 probe blocks DNA binding, whereas the SOX10 ligand increases SOX10-SOX10 interactions and disrupts melanoma transcriptional signaling. Our findings reveal heterogeneity in cysteine ligandability across cancers, pinpoint cell-intrinsic features driving cysteine targeting, and illustrate the use of covalent probes to disrupt oncogenic transcription-factor activity.

MolProphet: A One-Stop, General Purpose, and AI-Based Platform for the Early Stages of Drug Discovery

Artificial intelligence (AI) is an effective tool to accelerate drug discovery and cut costs in discovery processes. Many successful AI applications are reported in the early stages of small molecule drug discovery. However, most of those applications require a deep understanding of software and hardware, and focus on a single field that implies data normalization and transfer between those applications is still a challenge for normal users. It usually limits the application of AI in drug discovery. Here, based on a series of robust models, we formed a one-stop, general purpose, and AI-based drug discovery platform, MolProphet, to provide complete functionalities in the early stages of small molecule drug discovery, including AI-based target pocket prediction, hit discovery and lead optimization, and compound targeting, as well as abundant analyzing tools to check the results. MolProphet is an accessible and user-friendly web-based platform that is fully designed according to the practices in the drug discovery industry. The molecule screened, generated, or optimized by the MolProphet is purchasable and synthesizable at low cost but with good drug-likeness. More than 400 users from industry and academia have used MolProphet in their work. We hope this platform can provide a powerful solution to assist each normal researcher in drug design and related research areas. It is available for everyone at https://www.molprophet.com/.

Graph Attention Site Prediction (GrASP): Identifying Druggable Binding Sites Using Graph Neural Networks with Attention

Identifying and discovering druggable protein binding sites is an important early step in computer-aided drug discovery, but it remains a difficult task where most campaigns rely on a priori knowledge of binding sites from experiments. Here, we present a binding site prediction method called Graph Attention Site Prediction (GrASP) and re-evaluate assumptions in nearly every step in the site prediction workflow from data set preparation to model evaluation. GrASP is able to achieve state-of-the-art performance at recovering binding sites in PDB structures while maintaining a high degree of precision which will minimize wasted computation in downstream tasks such as docking and free energy perturbation.

Deep learning in structural bioinformatics: current applications and future perspectives

In this review article, we explore the transformative impact of deep learning (DL) on structural bioinformatics, emphasizing its pivotal role in a scientific revolution driven by extensive data, accessible toolkits and robust computing resources. As big data continue to advance, DL is poised to become an integral component in healthcare and biology, revolutionizing analytical processes. Our comprehensive review provides detailed insights into DL, featuring specific demonstrations of its notable applications in bioinformatics. We address challenges tailored for DL, spotlight recent successes in structural bioinformatics and present a clear exposition of DL-from basic shallow neural networks to advanced models such as convolution, recurrent, artificial and transformer neural networks. This paper discusses the emerging use of DL for understanding biomolecular structures, anticipating ongoing developments and applications in the realm of structural bioinformatics.

Learnt representations of proteins can be used for accurate prediction of small molecule binding sites on experimentally determined and predicted protein structures

Protein-ligand binding site prediction is a useful tool for understanding the functional behaviour and potential drug-target interactions of a novel protein of interest. However, most binding site prediction methods are tested by providing crystallised ligand-bound (holo) structures as input. This testing regime is insufficient to understand the performance on novel protein targets where experimental structures are not available. An alternative option is to provide computationally predicted protein structures, but this is not commonly tested. However, due to the training data used, computationally-predicted protein structures tend to be extremely accurate, and are often biased toward a holo conformation. In this study we describe and benchmark IF-SitePred, a protein-ligand binding site prediction method which is based on the labelling of ESM-IF1 protein language model embeddings combined with point cloud annotation and clustering. We show that not only is IF-SitePred competitive with state-of-the-art methods when predicting binding sites on experimental structures, but it performs better on proxies for novel proteins where low accuracy has been simulated by molecular dynamics. Finally, IF-SitePred outperforms other methods if ensembles of predicted protein structures are generated.

PLAS-20k: Extended Dataset of Protein-Ligand Affinities from MD Simulations for Machine Learning Applications

Computing binding affinities is of great importance in drug discovery pipeline and its prediction using advanced machine learning methods still remains a major challenge as the existing datasets and models do not consider the dynamic features of protein-ligand interactions. To this end, we have developed PLAS-20k dataset, an extension of previously developed PLAS-5k, with 97,500 independent simulations on a total of 19,500 different protein-ligand complexes. Our results show good correlation with the available experimental values, performing better than docking scores. This holds true even for a subset of ligands that follows Lipinski's rule, and for diverse clusters of complex structures, thereby highlighting the importance of PLAS-20k dataset in developing new ML models. Along with this, our dataset is also beneficial in classifying strong and weak binders compared to docking. Further, OnionNet model has been retrained on PLAS-20k dataset and is provided as a baseline for the prediction of binding affinities. We believe that large-scale MD-based datasets along with trajectories will form new synergy, paving the way for accelerating drug discovery.

Unraveling viral drug targets: a deep learning-based approach for the identification of potential binding sites

The coronavirus disease 2019 (COVID-19) pandemic has spurred a wide range of approaches to control and combat the disease. However, selecting an effective antiviral drug target remains a time-consuming challenge. Computational methods offer a promising solution by efficiently reducing the number of candidates. In this study, we propose a structure- and deep learning-based approach that identifies vulnerable regions in viral proteins corresponding to drug binding sites. Our approach takes into account the protein dynamics, accessibility and mutability of the binding site and the putative mechanism of action of the drug. We applied this technique to validate drug targeting toward severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike glycoprotein S. Our findings reveal a conformation- and oligomer-specific glycan-free binding site proximal to the receptor binding domain. This site comprises topologically important amino acid residues. Molecular dynamics simulations of Spike in complex with candidate drug molecules bound to the potential binding sites indicate an equilibrium shifted toward the inactive conformation compared with drug-free simulations. Small molecules targeting this binding site have the potential to prevent the closed-to-open conformational transition of Spike, thereby allosterically inhibiting its interaction with human angiotensin-converting enzyme 2 receptor. Using a pseudotyped virus-based assay with a SARS-CoV-2 neutralizing antibody, we identified a set of hit compounds that exhibited inhibition at micromolar concentrations.

Integrated Molecular Modeling and Machine Learning for Drug Design

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

DrugMap: A quantitative pan-cancer analysis of cysteine ligandability

Cysteine-focused chemical proteomic platforms have accelerated the clinical development of covalent inhibitors of a wide-range of targets in cancer. However, how different oncogenic contexts influence cysteine targeting remains unknown. To address this question, we have developed DrugMap , an atlas of cysteine ligandability compiled across 416 cancer cell lines. We unexpectedly find that cysteine ligandability varies across cancer cell lines, and we attribute this to differences in cellular redox states, protein conformational changes, and genetic mutations. Leveraging these findings, we identify actionable cysteines in NFκB1 and SOX10 and develop corresponding covalent ligands that block the activity of these transcription factors. We demonstrate that the NFκB1 probe blocks DNA binding, whereas the SOX10 ligand increases SOX10-SOX10 interactions and disrupts melanoma transcriptional signaling. Our findings reveal heterogeneity in cysteine ligandability across cancers, pinpoint cell-intrinsic features driving cysteine targeting, and illustrate the use of covalent probes to disrupt oncogenic transcription factor activity.

RefinePocket: An Attention-Enhanced and Mask-Guided Deep Learning Approach for Protein Binding Site Prediction

Protein binding site prediction is an important prerequisite task of drug discovery and design. While binding sites are very small, irregular and varied in shape, making the prediction very challenging. Standard 3D U-Net has been adopted to predict binding sites but got stuck with unsatisfactory prediction results, incomplete, out-of-bounds, or even failed. The reason is that this scheme is less capable of extracting the chemical interactions of the entire region and hardly takes into account the difficulty of segmenting complex shapes. In this paper, we propose a refined U-Net architecture, called RefinePocket, consisting of an attention-enhanced encoder and a mask-guided decoder. During encoding, taking binding site proposal as input, we employ Dual Attention Block (DAB) hierarchically to capture rich global information, exploring residue relationship and chemical correlations in spatial and channel dimensions respectively. Then, based on the enhanced representation extracted by the encoder, we devise Refine Block (RB) in the decoder to enable self-guided refinement of uncertain regions gradually, resulting in more precise segmentation. Experiments show that DAB and RB complement and promote each other, making RefinePocket has an average improvement of 10.02% on DCC and 4.26% on DVO compared with the state-of-the-art method on four test sets.

StackCPA: A stacking model for compound-protein binding affinity prediction based on pocket multi-scale features

Accurately predicting compound-protein binding affinity is a crucial task in drug discovery. Computational models offer the advantages of short time, low cost and safety compared to traditional drug development. Pocket is the key binding region of the protein, which provides invaluable information for drug repositioning and drug design. In this study, we propose an ensemble learning model, called StackCPA, to predict the compound-protein binding affinity. The model integrates multi-scale features of protein pocket and compound through a transfer learning strategy. The protein pocket is described in a fine-grained way by atomic level, residue level and subdomain level. The proposed model StackCPA is evaluated on three binding affinity benchmark datasets. The experiment results show that StackCPA achieves the best performance on all the three datasets in comparison with other state-of-the-art deep learning models. The ablation study shows that the protein pocket can provide sufficient information for affinity prediction and its multi-scale features enable the model to further improve the prediction performance. In addition, the case study for epidermal growth factor receptor erbB1 (EGFR) indicates that StackCPA could serve as an effective tool for drug repurposing. Source codes and data of StackCPA are available at https://github.com/CSUBioGroup/StackCPA.

Open-Source Machine Learning in Computational Chemistry

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

LigBind: identifying binding residues for over 1000 ligands with relation-aware graph neural networks

Identifying the interactions between proteins and ligands is significant for drug discovery and design. Considering the diverse binding patterns of ligands, the ligand-specific methods are trained per ligand to predict binding residues. However, most of the existing ligand-specific methods ignore shared binding preferences among various ligands and generally only cover a limited number of ligands with a sufficient number of known binding proteins. In this study, we propose a relation-aware framework LigBind with graph-level pre-training to enhance the ligand-specific binding residue predictions for 1159 ligands, which can effectively cover the ligands with a few known binding proteins. LigBind first pre-trains a graph neural network-based feature extractor for ligand-residue pairs and relation-aware classifiers for similar ligands. Then, LigBind is fine-tuned with ligand-specific binding data, where a domain adaptive neural network is designed to automatically leverage the diversity and similarity of various ligand-binding patterns for accurate binding residue prediction. We construct ligand-specific benchmark datasets of 1159 ligands and 16 unseen ligands, which are used to evaluate the effectiveness of LigBind. The results demonstrate the LigBind's efficacy on the large-scale ligand-specific benchmark datasets, and generalizes well to unseen ligands. LigBind also enables accurate identification of the ligand-binding residues in the main protease, papain-like protease and the RNA-dependent RNA polymerase of SARS-CoV-2. The webserver and source codes of LigBind are available at http://www.csbio.sjtu.edu.cn/bioinf/LigBind/ and https://github.com/YYingXia/LigBind/ for academic use.

Advancing Targeted Protein Degradation via Multiomics Profiling and Artificial Intelligence

Only around 20% of the human proteome is considered to be druggable with small-molecule antagonists. This leaves some of the most compelling therapeutic targets outside the reach of ligand discovery. The concept of targeted protein degradation (TPD) promises to overcome some of these limitations. In brief, TPD is dependent on small molecules that induce the proximity between a protein of interest (POI) and an E3 ubiquitin ligase, causing ubiquitination and degradation of the POI. In this perspective, we want to reflect on current challenges in the field, and discuss how advances in multiomics profiling, artificial intelligence, and machine learning (AI/ML) will be vital in overcoming them. The presented roadmap is discussed in the context of small-molecule degraders but is equally applicable for other emerging proximity-inducing modalities.

CAT-Site: Predicting Protein Binding Sites Using a Convolutional Neural Network

Identifying binding sites on the protein surface is an important part of computer-assisted drug design processes. Reliable prediction of binding sites not only assists with docking algorithms, but it can also explain the possible side-effects of a potential drug as well as its efficiency. In this work, we propose a novel workflow for predicting possible binding sites of a ligand on a protein surface. We use proteins from the PDBbind and sc-PDB databases, from which we combine available ligand information for similar proteins using all the possible ligands rather than only a special sub-selection to generalize the work of existing research. After performing protein clustering and merging of ligands of similar proteins, we use a three-dimensional convolutional neural network that takes into account the spatial structure of a protein. Lastly, we combine ligandability predictions for points on protein surfaces into joint binding sites. Analysis of our model's performance shows that its achieved sensitivity is 0.829, specificity is 0.98, and F1 score is 0.517, and that for 54% of larger and pharmacologically relevant binding sites, the distance between their real and predicted centers amounts to less than 4 Å.

BindWeb: A web server for ligand binding residue and pocket prediction from protein structures

Knowledge of protein-ligand interactions is beneficial for biological process analysis and drug design. Given the complexity of the interactions and the inadequacy of experimental data, accurate ligand binding residue and pocket prediction remains challenging. In this study, we introduce an easy-to-use web server BindWeb for ligand-specific and ligand-general binding residue and pocket prediction from protein structures. BindWeb integrates a graph neural network GraphBind with a hybrid convolutional neural network and bidirectional long short-term memory network DELIA to identify binding residues. Furthermore, BindWeb clusters the predicted binding residues to binding pockets with mean shift clustering. The experimental results and case study demonstrate that BindWeb benefits from the complementarity of two base methods. BindWeb is freely available for academic use at http://www.csbio.sjtu.edu.cn/bioinf/BindWeb/.

Holo Protein Conformation Generation from Apo Structures by Ligand Binding Site Refinement

An important part in structure-based drug design is the selection of an appropriate protein structure. It has been revealed that a holo protein structure that contains a well-defined binding site is a much better choice than an apo structure in structure-based drug discovery. Therefore, it is valuable to obtain a holo-like protein conformation from apo structures in the case where no holo structure is available. Meeting the need, we present a robust approach to generate reliable holo-like structures from apo structures by ligand binding site refinement with restraints derived from holo templates with low homology. Our method was tested on a test set of 32 proteins from the DUD-E data set and compared with other approaches. It was shown that our method successfully refined the apo structures toward the corresponding holo conformations for 23 of 32 proteins, reducing the average all-heavy-atom RMSD of binding site residues by 0.48 Å. In addition, when evaluated against all the holo structures in the protein data bank, our method can improve the binding site RMSD for 14 of 19 cases that experience significant conformational changes. Furthermore, our refined structures also demonstrate their advantages over the apo structures in ligand binding mode predictions by both rigid docking and flexible docking and in virtual screening on the database of active and decoy ligands from the DUD-E. These results indicate that our method is effective in recovering holo-like conformations and will be valuable in structure-based drug discovery.

Celastrol Targets Cullin-Associated and Neddylation-Dissociated 1 to Prevent Fibroblast-Myofibroblast Transformation against Pulmonary Fibrosis

Celastrol (CEL), a pentacyclic triterpene compound, has been proven to have a definite antipulmonary fibrosis effect. However, its direct targets for antipulmonary fibrosis remain unknown. In this study, we designed and synthesized a series of celastrol-based probes to identify the direct targets in human pulmonary fibroblasts using an activity-based protein profiling strategy. Among many fished targets, we identified a key protein, cullin-associated and neddylation-dissociated 1 (CAND1), which was involved in fibroblast-myofibroblast transformation (FMT). More importantly, we found that the inhibitory effect of celastrol on FMT is dependent on CAND1, through improving the interactions between CAND1 and Cullin1 to promote the activity of Skp1/Cullin1/F-box ubiquitin ligases. In silico studies and cysteine mutation experiments further demonstrated that Cys264 of CAND1 is the site for conjugation of celastrol. This reveals a new mechanism of celastrol against pulmonary fibrosis and may provide a novel therapeutic option for antipulmonary fibrosis.

Estimating the Similarity between Protein Pockets

With the exponential increase in publicly available protein structures, the comparison of protein binding sites naturally emerged as a scientific topic to explain observations or generate hypotheses for ligand design, notably to predict ligand selectivity for on- and off-targets, explain polypharmacology, and design target-focused libraries. The current review summarizes the state-of-the-art computational methods applied to pocket detection and comparison as well as structural druggability estimates. The major strengths and weaknesses of current pocket descriptors, alignment methods, and similarity search algorithms are presented. Lastly, an exhaustive survey of both retrospective and prospective applications in diverse medicinal chemistry scenarios illustrates the capability of the existing methods and the hurdle that still needs to be overcome for more accurate predictions.

The druggable genome: Twenty years later

The concept of the druggable genome has been with us for 20 years. During this time, researchers have developed several methods and resources to help assess a target’s druggability. In parallel, evidence for target-disease associations has been collated at scale by Open Targets. More recently, the Protein Data Bank in Europe (PDBe) have built a knowledge base matching per-residue annotations with available protein structure. While each resource is useful in isolation, we believe there is enormous potential in bringing all relevant data into a single knowledge graph, from gene-level to protein residue. Automation is vital for the processing and assessment of all available structures. We have developed scalable, automated workflows that provide hotspot-based druggability assessments for all available structures across large numbers of targets. Ultimately, we will run our method at a proteome scale, an ambition made more realistic by the arrival of AlphaFold 2. Bringing together annotations from the residue up to the gene level and building connections within the graph to represent pathways or protein-protein interactions will create complexity that mirrors the biological systems they represent. Such complexity is difficult for the human mind to utilise effectively, particularly at scale. We believe that graph-based AI methods will be able to expertly navigate such a knowledge graph, selecting the targets of the future.

A reinforcement learning approach for protein-ligand binding pose prediction

Protein ligand docking is an indispensable tool for computational prediction of protein functions and screening drug candidates. Despite significant progress over the past two decades, it is still a challenging problem, characterized by the still limited understanding of the energetics between proteins and ligands, and the vast conformational space that has to be searched to find a satisfactory solution. In this project, we developed a novel reinforcement learning (RL) approach, the asynchronous advantage actor-critic model (A3C), to address the protein ligand docking problem. The overall framework consists of two models. During the search process, the agent takes an action selected by the actor model based on the current location. The critic model then evaluates this action and predict the distance between the current location and true binding site. Experimental results showed that in both single- and multi-atom cases, our model improves binding site prediction substantially compared to a naïve model. For the single-atom ligand, copper ion (Cu²⁺), the model predicted binding sites have a median root-mean-square-deviation (RMSD) of 2.39 Å to the true binding sites when starting from random starting locations. For the multi-atom ligand, sulfate ion (SO₄^2-), the predicted binding sites have a median RMSD of 3.82 Å to the true binding sites. The ligand-specific models built in this study can be used in solvent mapping studies and the RL framework can be readily scaled up to larger and more diverse sets of ligands.

Probabilistic Pocket Druggability Prediction via One-Class Learning

The choice of target pocket is a key step in a drug discovery campaign. This step can be supported by in silico druggability prediction. In the literature, druggability prediction is often approached as a two-class classification task that distinguishes between druggable and non-druggable (or less druggable) pockets (or voxels). Apart from obvious cases, however, the non-druggable class is conceptually ambiguous. This is because any pocket (or target) is only non-druggable until a drug is found for it. It is therefore more appropriate to adopt a one-class approach, which uses only unambiguous information, namely, druggable pockets. Here, we propose using the import vector domain description (IVDD) algorithm to support this task. IVDD is a one-class probabilistic kernel machine that we previously introduced. To feed the algorithm, we use customized DrugPred descriptors computed via NanoShaper. Our results demonstrate the feasibility and effectiveness of the approach. In particular, we can remove or mitigate biases chiefly due to the labeling.

Prediction of allosteric druggable pockets of cyclin-dependent kinases

Cyclin-dependent kinase (Cdk) proteins play crucial roles in the cell cycle progression and are thus attractive drug targets for therapy against such aberrant cell cycle processes as cancer. Since most of the available Cdk inhibitors target the highly conserved catalytic ATP pocket and their lack of specificity often lead to side effects, it is imperative to identify and characterize less conserved non-catalytic pockets capable of interfering with the kinase activity allosterically. However, a systematic analysis of these allosteric druggable pockets is still in its infancy. Here, we summarize the existing Cdk pockets and their selectivity. Then, we outline a network-based pocket prediction approach (NetPocket) and illustrate its utility for systematically identifying the allosteric druggable pockets with case studies. Finally, we discuss potential future directions and their challenges.

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.

BiRDS - Binding Residue Detection from Protein Sequences Using Deep ResNets

Protein-drug interactions play important roles in many biological processes and therapeutics. Predicting the binding sites of a protein helps to discover such interactions. New drugs can be designed to optimize these interactions, improving protein function. The tertiary structure of a protein decides the binding sites available to the drug molecule, but the determination of the 3D structure is slow and expensive. Conversely, the determination of the amino acid sequence is swift and economical. Although quick and accurate prediction of the binding site using just the sequence is challenging, the application of Deep Learning, which has been hugely successful in several biochemical tasks, makes it feasible. BiRDS is a Residual Neural Network that predicts the protein's most active binding site using sequence information. SC-PDB, an annotated database of druggable binding sites, is used for training the network. Multiple Sequence Alignments of the proteins in the database are generated using DeepMSA, and features such as Position-Specific Scoring Matrix, Secondary Structure, and Relative Solvent Accessibility are extracted. During training, a weighted binary cross-entropy loss function is used to counter the substantial imbalance in the two classes of binding and nonbinding residues. A novel test set SC6K is introduced to compare binding-site prediction methods. BiRDS achieves an AUROC score of 0.87, and the center of 25% of its predicted binding sites lie within 4 Å of the center of the actual binding site.

3DLigandSite: structure-based prediction of protein-ligand binding sites

3DLigandSite is a web tool for the prediction of ligand-binding sites in proteins. Here, we report a significant update since the first release of 3DLigandSite in 2010. The overall methodology remains the same, with candidate binding sites in proteins inferred using known binding sites in related protein structures as templates. However, the initial structural modelling step now uses the newly available structures from the AlphaFold database or alternatively Phyre2 when AlphaFold structures are not available. Further, a sequence-based search using HHSearch has been introduced to identify template structures with bound ligands that are used to infer the ligand-binding residues in the query protein. Finally, we introduced a machine learning element as the final prediction step, which improves the accuracy of predictions and provides a confidence score for each residue predicted to be part of a binding site. Validation of 3DLigandSite on a set of 6416 binding sites obtained 92% recall at 75% precision for non-metal binding sites and 52% recall at 75% precision for metal binding sites. 3DLigandSite is available at https://www.wass-michaelislab.org/3dligandsite. Users submit either a protein sequence or structure. Results are displayed in multiple formats including an interactive Mol* molecular visualization of the protein and the predicted binding sites.

Modern Machine Learning for Tackling Inverse Problems in Chemistry: Molecular Design to Realization

The discovery of new molecules and materials helps expand the horizons of novel and innovative real-life applications. In the pursuit of finding molecules with desired properties, chemists have traditionally relied...

MoleGuLAR: Molecule Generation Using Reinforcement Learning with Alternating Rewards

The design of new inhibitors for novel targets is a very important problem especially in the current scenario with the world being plagued by COVID-19. Conventional approaches such as high-throughput virtual screening require extensive combing through existing data sets in the hope of finding possible matches. In this study, we propose a computational strategy for de novo generation of molecules with high binding affinities to the specified target and other desirable properties for druglike molecules using reinforcement learning. A deep generative model built using a stack-augmented recurrent neural network initially trained to generate druglike molecules is optimized using reinforcement learning to start generating molecules with desirable properties like LogP, quantitative estimate of drug likeliness, topological polar surface area, and hydration free energy along with the binding affinity. For multiobjective optimization, we have devised a novel strategy in which the property being used to calculate the reward is changed periodically. In comparison to the conventional approach of taking a weighted sum of all rewards, this strategy shows an enhanced ability to generate a significantly higher number of molecules with desirable properties.

Artificial intelligence: machine learning for chemical sciences

Research in molecular sciences witnessed the rise and fall of Artificial Intelligence (AI)/ Machine Learning (ML) methods, especially artificial neural networks, few decades ago. However, we see a major resurgence in the use of modern ML methods in scientific research during the last few years. These methods have had phenomenal success in the areas of computer vision, speech recognition, natural language processing (NLP), etc. This has inspired chemists and biologists to apply these algorithms to problems in natural sciences. Availability of high performance Graphics Processing Unit (GPU) accelerators, large datasets, new algorithms, and libraries has enabled this surge. ML algorithms have successfully been applied to various domains in molecular sciences by providing much faster and sometimes more accurate solutions compared to traditional methods like Quantum Mechanical (QM) calculations, Density Functional Theory (DFT) or Molecular Mechanics (MM) based methods, etc. Some of the areas where the potential of ML methods are shown to be effective are in drug design, prediction of high-level quantum mechanical energies, molecular design, molecular dynamics materials, and retrosynthesis of organic compounds, etc. This article intends to conceptually introduce various modern ML methods and their relevance and applications in computational natural sciences. Synopsis Recent surge in the application of machine learning (ML) methods in fundamental sciences has led to a perspective that these methods may become important tools in chemical science. This perspective provides an overview of the modern ML methods and their successful applications in chemistry during the last few years.

SCONES: Self-Consistent Neural Network for Protein Stability Prediction Upon Mutation

Engineering proteins to have desired properties by mutating amino acids at specific sites is commonplace. Such engineered proteins must be stable to function. Experimental methods used to determine stability at throughputs required to scan the protein sequence space thoroughly are laborious. To this end, many machine learning based methods have been developed to predict thermodynamic stability changes upon mutation. These methods have been evaluated for symmetric consistency by testing with hypothetical reverse mutations. In this work, we propose transitive data augmentation, evaluating transitive consistency with our new S^transitive data set, and a new machine learning based method, the first of its kind, that incorporates both symmetric and transitive properties into the architecture. Our method, called SCONES, is an interpretable neural network that predicts small relative protein stability changes for missense mutations that do not significantly alter the structure. It estimates a residue's contributions toward protein stability (ΔG) in its local structural environment, and the difference between independently predicted contributions of the reference and mutant residues is reported as ΔΔG. We show that this self-consistent machine learning architecture is immune to many common biases in data sets, relies less on data than existing methods, is robust to overfitting, and can explain a substantial portion of the variance in experimental data.

MEMES: Machine learning framework for Enhanced MolEcular Screening

In drug discovery applications, high throughput virtual screening exercises are routinely performed to determine an initial set of candidate molecules referred to as "hits". In such an experiment, each molecule from a large small-molecule drug library is evaluated in terms of physical properties such as the docking score against a target receptor. In real-life drug discovery experiments, drug libraries are extremely large but still there is only a minor representation of the essentially infinite chemical space, and evaluation of physical properties for each molecule in the library is not computationally feasible. In the current study, a novel Machine learning framework for Enhanced MolEcular Screening (MEMES) based on Bayesian optimization is proposed for efficient sampling of the chemical space. The proposed framework is demonstrated to identify 90% of the top-1000 molecules from a molecular library of size about 100 million, while calculating the docking score only for about 6% of the complete library. We believe that such a framework would tremendously help to reduce the computational effort in not only drug-discovery but also areas that require such high-throughput experiments.

PUResNet: prediction of protein-ligand binding sites using deep residual neural network

Background

Predicting protein-ligand binding sites is a fundamental step in understanding the functional characteristics of proteins, which plays a vital role in elucidating different biological functions and is a crucial step in drug discovery. A protein exhibits its true nature after binding to its interacting molecule known as a ligand that binds only in the favorable binding site of the protein structure. Different computational methods exploiting the features of proteins have been developed to identify the binding sites in the protein structure, but none seems to provide promising results, and therefore, further investigation is required.

Results

In this study, we present a deep learning model PUResNet and a novel data cleaning process based on structural similarity for predicting protein-ligand binding sites. From the whole scPDB (an annotated database of druggable binding sites extracted from the Protein DataBank) database, 5020 protein structures were selected to address this problem, which were used to train PUResNet. With this, we achieved better and justifiable performance than the existing methods while evaluating two independent sets using distance, volume and proportion metrics.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13321-021-00547-7.