MOTIVATION: Fast and accurate classification of ligand-binding sites in proteins with respect to the class of binding molecules is invaluable not only to the automatic functional annotation of large datasets of protein structures but also to projects in protein evolution, protein engineering and drug development. Deep learning techniques, which have already been successfully applied to address challenging problems across various fields, are inherently suitable to classify ligand-binding pockets. Our goal is to demonstrate that off-the-shelf deep learning models can be employed with minimum development effort to recognize nucleotide- and heme-binding sites with a comparable accuracy to highly specialized, voxel-based methods. RESULTS: We developed BionoiNet, a new deep learning-based framework implementing a popular ResNet model for image classification. BionoiNet first transforms the molecular structures of ligand-binding sites to 2D Voronoi diagrams, which are then used as the input to a pretrained convolutional neural network classifier. The ResNet model generalizes well to unseen data achieving the accuracy of 85.6% for nucleotide- and 91.3% for heme-binding pockets. BionoiNet also computes significance scores of pocket atoms, called BionoiScores, to provide meaningful insights into their interactions with ligand molecules. BionoiNet is a lightweight alternative to computationally expensive 3D architectures. AVAILABILITY AND IMPLEMENTATION: BionoiNet is implemented in Python with the source code freely available at: https://github.com/CSBG-LSU/BionoiNet. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Interactions between proteins and small molecules are critical for biological functions. These interactions often occur in small cavities within protein structures, known as ligand-binding pockets. Understanding the physicochemical qualities of binding pockets is essential to improve not only our basic knowledge of biological systems, but also drug development procedures. In order to quantify similarities among pockets in terms of their geometries and chemical properties, either bound ligands can be compared to one another or binding sites can be matched directly. Both perspectives routinely take advantage of computational methods including various techniques to represent and compare small molecules as well as local protein structures. In this review, we survey 12 tools widely used to match pockets. These methods are divided into five categories based on the algorithm implemented to construct binding-site alignments. In addition to the comprehensive analysis of their algorithms, test sets and the performance of each method are described. We also discuss general pharmacological applications of computational pocket matching in drug repurposing, polypharmacology and side effects. Reflecting on the importance of these techniques in drug discovery, in the end, we elaborate on the development of more accurate meta-predictors, the incorporation of protein flexibility and the integration of powerful artificial intelligence technologies such as deep learning.

Comprehensive characterization of ligand-binding sites is invaluable to infer molecular functions of hypothetical proteins, trace evolutionary relationships between proteins, engineer enzymes to achieve a desired substrate specificity, and develop drugs with improved selectivity profiles. These research efforts pose significant challenges owing to the fact that similar pockets are commonly observed across different folds, leading to the high degree of promiscuity of ligand-protein interactions at the system-level. On that account, novel algorithms to accurately classify binding sites are needed. Deep learning is attracting a significant attention due to its successful applications in a wide range of disciplines. In this communication, we present DeepDrug3D, a new approach to characterize and classify binding pockets in proteins with deep learning. It employs a state-of-the-art convolutional neural network in which biomolecular structures are represented as voxels assigned interaction energy-based attributes. The current implementation of DeepDrug3D, trained to detect and classify nucleotide- and heme-binding sites, not only achieves a high accuracy of 95%, but also has the ability to generalize to unseen data as demonstrated for steroid-binding proteins and peptidase enzymes. Interestingly, the analysis of strongly discriminative regions of binding pockets reveals that this high classification accuracy arises from learning the patterns of specific molecular interactions, such as hydrogen bonds, aromatic and hydrophobic contacts. DeepDrug3D is available as an open-source program at https://github.com/pulimeng/DeepDrug3D with the accompanying TOUGH-C1 benchmarking dataset accessible from https://osf.io/enz69/.

BACKGROUND: Detecting similar ligand-binding sites in globally unrelated proteins has a wide range of applications in modern drug discovery, including drug repurposing, the prediction of side effects, and drug-target interactions. Although a number of techniques to compare binding pockets have been developed, this problem still poses significant challenges. RESULTS: We evaluate the performance of three algorithms to calculate similarities between ligand-binding sites, APoc, SiteEngine, and G-LoSA. Our assessment considers not only the capabilities to identify similar pockets and to construct accurate local alignments, but also the dependence of these alignments on the sequence order. We point out certain drawbacks of previously compiled datasets, such as the inclusion of structurally similar proteins, leading to an overestimated performance. To address these issues, a rigorous procedure to prepare unbiased, high-quality benchmarking sets is proposed. Further, we conduct a comparative assessment of techniques directly aligning binding pockets to indirect strategies employing structure-based virtual screening with AutoDock Vina and rDock. CONCLUSIONS: Thorough benchmarks reveal that G-LoSA offers a fairly robust overall performance, whereas the accuracy of APoc and SiteEngine is satisfactory only against easy datasets. Moreover, combining various algorithms into a meta-predictor improves the performance of existing methods to detect similar binding sites in unrelated proteins by 5-10%. All data reported in this paper are freely available at https://osf.io/6ngbs/ .

BACKGROUND: Deciphering complete networks of interactions between proteins is the key to comprehend cellular regulatory mechanisms. A significant effort has been devoted to expanding the coverage of the proteome-wide interaction space at molecular level. Although a growing body of research shows that protein docking can, in principle, be used to predict biologically relevant interactions, the accuracy of the across-proteome identification of interacting partners and the selection of near-native complex structures still need to be improved. RESULTS: In this study, we developed a new method to discover and model protein interactions employing an exhaustive all-to-all docking strategy. This approach integrates molecular modeling, structural bioinformatics, machine learning, and functional annotation filters in order to provide interaction data for the bottom-up assembly of protein interaction networks. Encouragingly, the success rates for dimer modeling is 57.5 and 48.7% when experimental and computer-generated monomer structures are employed, respectively. Further, our protocol correctly identifies 81% of protein-protein interactions at the expense of only 19% false positive rate. As a proof of concept, 61,913 protein-protein interactions were confidently predicted and modeled for the proteome of E. coli. Finally, we validated our method against the human immune disease pathway. CONCLUSIONS: Protein docking supported by evolutionary restraints and machine learning can be used to reliably identify and model biologically relevant protein assemblies at the proteome scale. Moreover, the accuracy of the identification of protein-protein interactions is improved by considering only those protein pairs co-localized in the same cellular compartment and involved in the same biological process. The modeling protocol described in this communication can be applied to detect protein-protein interactions in other organisms and pathways as well as to construct dimer structures and estimate the confidence of protein interactions experimentally identified with high-throughput techniques.

MOTIVATION: Many drugs interact with numerous proteins besides their intended therapeutic targets and a substantial portion of these interactions is yet to be elucidated. Protein-Drug Interaction Database (PDID) addresses incompleteness of these data by providing access to putative protein-drug interactions that cover the entire structural human proteome. RESULTS: PDID covers 9652 structures from 3746 proteins and houses 16 800 putative interactions generated from close to 1.1 million accurate, all-atom structure-based predictions for several dozens of popular drugs. The predictions were generated with three modern methods: ILbind, SMAP and eFindSite. They are accompanied by propensity scores that quantify likelihood of interactions and coordinates of the putative location of the binding drugs in the corresponding protein structures. PDID complements the current databases that focus on the curated interactions and the BioDrugScreen database that relies on docking to find putative interactions. Moreover, we also include experimentally curated interactions which are linked to their sources: DrugBank, BindingDB and Protein Data Bank. Our database can be used to facilitate studies related to polypharmacology of drugs including repurposing and explaining side effects of drugs. AVAILABILITY AND IMPLEMENTATION: PDID database is freely available at http://biomine.ece.ualberta.ca/PDID/.

It has been more than a decade since the completion of the Human Genome Project that provided us with a complete list of human proteins. The next obvious task is to figure out how various parts interact with each other. On that account, we review 10 methods for protein interface prediction, which are freely available as web servers. In addition, we comparatively evaluate their performance on a common data set comprising different quality target structures. We find that using experimental structures and high-quality homology models, structure-based methods outperform those using only protein sequences, with global template-based approaches providing the best performance. For moderate-quality models, sequence-based methods often perform better than those structure-based techniques that rely on fine atomic details. We note that post-processing protocols implemented in several methods quantitatively improve the results only for experimental structures, suggesting that these procedures should be tuned up for computer-generated models. Finally, we anticipate that advanced meta-prediction protocols are likely to enhance interface residue prediction. Notwithstanding further improvements, easily accessible web servers already provide the scientific community with convenient resources for the identification of protein-protein interaction sites.

Detecting similarities between ligand binding sites in the absence of global homology between target proteins has been recognized as one of the critical components of modern drug discovery. Local binding site alignments can be constructed using sequence order-independent techniques, however, to achieve a high accuracy, many current algorithms for binding site comparison require high-quality experimental protein structures, preferably in the bound conformational state. This, in turn, complicates proteome scale applications, where only various quality structure models are available for the majority of gene products. To improve the state-of-the-art, we developed eMatchSite, a new method for constructing sequence order-independent alignments of ligand binding sites in protein models. Large-scale benchmarking calculations using adenine-binding pockets in crystal structures demonstrate that eMatchSite generates accurate alignments for almost three times more protein pairs than SOIPPA. More importantly, eMatchSite offers a high tolerance to structural distortions in ligand binding regions in protein models. For example, the percentage of correctly aligned pairs of adenine-binding sites in weakly homologous protein models is only 4-9% lower than those aligned using crystal structures. This represents a significant improvement over other algorithms, e.g. the performance of eMatchSite in recognizing similar binding sites is 6% and 13% higher than that of SiteEngine using high- and moderate-quality protein models, respectively. Constructing biologically correct alignments using predicted ligand binding sites in protein models opens up the possibility to investigate drug-protein interaction networks for complete proteomes with prospective systems-level applications in polypharmacology and rational drug repositioning. eMatchSite is freely available to the academic community as a web-server and a stand-alone software distribution at http://www.brylinski.org/ematchsite.

UNLABELLED: In the post-genomic era, the annotation of protein function facilitates the understanding of various biological processes. To extend the range of function annotation methods to the twilight zone of sequence identity, we have developed approaches that exploit both protein tertiary structure and/or protein sequence evolutionary relationships. To serve the scientific community, we have integrated the structure prediction tools, TASSER, TASSER-Lite and METATASSER, and the functional inference tools, FINDSITE, a structure-based algorithm for binding site prediction, Gene Ontology molecular function inference and ligand screening, EFICAz(2), a sequence-based approach to enzyme function inference and DBD-hunter, an algorithm for predicting DNA-binding proteins and associated DNA-binding residues, into a unified web resource, Protein Structure and Function prediction Resource (PSiFR). AVAILABILITY AND IMPLEMENTATION: PSiFR is freely available for use on the web at http://psifr.cssb.biology.gatech.edu/

A key challenge of the post-genomic era is the identification of the function(s) of all the molecules in a given organism. Here, we review the status of sequence and structure-based approaches to protein function inference and ligand screening that can provide functional insights for a significant fraction of the approximately 50% of ORFs of unassigned function in an average proteome. We then describe FINDSITE, a recently developed algorithm for ligand binding site prediction, ligand screening and molecular function prediction, which is based on binding site conservation across evolutionary distant proteins identified by threading. Importantly, FINDSITE gives comparable results when high-resolution experimental structures as well as predicted protein models are used.

Ligand virtual screening is a widely used tool to assist in new pharmaceutical discovery. In practice, virtual screening approaches have a number of limitations, and the development of new methodologies is required. Previously, we showed that remotely related proteins identified by threading often share a common binding site occupied by chemically similar ligands. Here, we demonstrate that across an evolutionarily related, but distant family of proteins, the ligands that bind to the common binding site contain a set of strongly conserved anchor functional groups as well as a variable region that accounts for their binding specificity. Furthermore, the sequence and structure conservation of residues contacting the anchor functional groups is significantly higher than those contacting ligand variable regions. Exploiting these insights, we developed FINDSITE(LHM) that employs structural information extracted from weakly related proteins to perform rapid ligand docking by homology modeling. In large scale benchmarking, using the predicted anchor-binding mode and the crystal structure of the receptor, FINDSITE(LHM) outperforms classical docking approaches with an average ligand RMSD from native of approximately 2.5 A. For weakly homologous receptor protein models, using FINDSITE(LHM), the fraction of recovered binding residues and specific contacts is 0.66 (0.55) and 0.49 (0.38) for highly confident (all) targets, respectively. Finally, in virtual screening for HIV-1 protease inhibitors, using similarity to the ligand anchor region yields significantly improved enrichment factors. Thus, the rather accurate, computationally inexpensive FINDSITE(LHM) algorithm should be a useful approach to assist in the discovery of novel biopharmaceuticals.

A description of many biological processes requires knowledge of the 3-D structure of proteins and, in particular, the defined active site responsible for biological function. Many proteins, the genes of which have been identified as the result of human genome sequencing, and which were synthesized experimentally, await identification of their biological activity. Currently used methods do not always yield satisfactory results, and new algorithms need to be developed to recognize the localization of active sites in proteins. This paper describes a computational model that can be used to identify potential areas that are able to interact with other molecules (ligands, substrates, inhibitors, etc.). The model for active site recognition is based on the analysis of hydrophobicity distribution in protein molecules. It is shown, based on the analyses of proteins with known biological activity and of proteins of unknown function, that the region of significantly irregular hydrophobicity distribution in proteins appears to be function related.

MOTIVATION: The problem of early-stage protein folding is critical for protein structure prediction. The model presented introduces a common definition of protein structures which may be treated as the possible in silico early-stage form of the polypeptide chain. Limitation of the conformational space to the ellipse path on the Ramachandran map was tested as a possible sub-space to represent the early-stage structure for simulation of protein folding. The proposed conformational sub-space was developed on the basis of the backbone conformation, with side-chain interactions excluded. RESULTS: The ellipse-path-limited conformation of BPTI was created using the criterion of shortest distance between Phi, Psi angles in native form of protein and the Phi, Psi angles belonging to the ellipse. No knots were observed in the structure created according to ellipse-path conformational sub-space. The energy minimization procedure applied to ellipse-path derived conformation directed structural changes toward the native form of the protein with SS-bonds system introduced to the procedure. AVAILABILITY: Program 'Ellipse' to create the ellipse-path derived structure available on request: [email protected]