Combinations of protein-chemical complex structures reveal new targets for established drugs

Deep learning identifies explainable reasoning paths of mechanism of action for drug repurposing from multilayer biological network

The discovery and repurposing of drugs require a deep understanding of the mechanism of drug action (MODA). Existing computational methods mainly model MODA with the protein-protein interaction (PPI) network. However, the molecular interactions of drugs in the human body are far beyond PPIs. Additionally, the lack of interpretability of these models hinders their practicability. We propose an interpretable deep learning-based path-reasoning framework (iDPath) for drug discovery and repurposing by capturing MODA on by far the most comprehensive multilayer biological network consisting of the complex high-dimensional molecular interactions between genes, proteins and chemicals. Experiments show that iDPath outperforms state-of-the-art machine learning methods on a general drug repurposing task. Further investigations demonstrate that iDPath can identify explicit critical paths that are consistent with clinical evidence. To demonstrate the practical value of iDPath, we apply it to the identification of potential drugs for treating prostate cancer and hypertension. Results show that iDPath can discover new FDA-approved drugs. This research provides a novel interpretable artificial intelligence perspective on drug discovery.

Identification of the Prognostic Signatures of Glioma With Different PTEN Status

The high-grade glioma is characterized by cell heterogeneity, gene mutations, and poor prognosis. The deletions and mutations of the tumor suppressor gene PTEN (5%-40%) in glioma patients are associated with worse survival and therapeutic resistance. Characterization of unique prognosis molecular signatures by PTEN status in glioma is still unclear. This study established a novel risk model, screened optimal prognostic signatures, and calculated the risk score for the individual glioma patients with different PTEN status. Screening results revealed fourteen independent prognostic gene signatures in PTEN-wt and three in the -50PTEN-mut subgroup. Moreover, we verified risk score as an independent prognostic factor significantly correlated with tumor malignancy. Due to the higher malignancy of the PTEN-mut gliomas, we explored the independent prognostic signatures (CLCF1, AEBP1, and OS9) for a potential therapeutic target in PTEN-mut glioma. We further separated IDH wild-type glioma patients into GBM and LGG to verify the therapeutic target along with PTEN status, notably, the above screened therapeutic targets are also significant prognostic genes in both IDH-wt/PTEN-mut GBM and LGG patients. We further identified the small molecule compound (+)-JQ1 binds to all three targets, indicating a potential therapy for PTEN-mut glioma. In sum, gene signatures and risk scores in the novel risk model facilitate glioma diagnosis, prognosis prediction, and treatment.

Identification of Target Associations for Polypharmacology from Analysis of Crystallographic Ligands of the Protein Data Bank

The design of a chemical entity that potently and selectively binds to a biological target of therapeutic relevance has dominated the scene of drug discovery so far. However, recent findings suggest that multitarget ligands may be endowed with superior efficacy and be less prone to drug resistance. The Protein Data Bank (PDB) provides experimentally validated structural information about targets and bound ligands. Therefore, it represents a valuable source of information to help identifying active sites, understanding pharmacophore requirements, designing novel ligands, and inferring structure-activity relationships. In this study, we performed a large-scale analysis of the PDB by integrating different ligand-based and structure-based approaches, with the aim of identifying promising target associations for polypharmacology based on reported crystal structure information. First, the 2D and 3D similarity profiles of the crystallographic ligands were evaluated using different ligand-based methods. Then, activity data of pairs of similar ligands binding to different targets were inspected by comparing structural information with bioactivity annotations reported in the ChEMBL, BindingDB, BindingMOAD, and PDBbind databases. Afterward, extensive docking screenings of ligands in the identified cross-targets were made in order to validate and refine the ligand-based results. Finally, the therapeutic relevance of the identified target combinations for polypharmacology was evaluated from comparison with information on therapeutic targets reported in the Therapeutic Target Database (TTD). The results led to the identification of several target associations with high therapeutic potential for polypharmacology.

Selection of protein conformations for structure-based polypharmacology studies

Several drugs exert their therapeutic effect through the modulation of multiple targets. Structure-based approaches hold great promise for identifying compounds with the desired polypharmacological profiles. These methods use knowledge of the protein binding sites to identify stereoelectronically complementary ligands. The selection of the most suitable protein conformations to be used in the design process is vital, especially for multitarget drug design in which the same ligand has to be accommodated in multiple binding pockets. Herein, we focus on currently available techniques for the selection of the most suitable protein conformations for multitarget drug design, compare the potential advantages and limitations of each method, and comment on how their combination could help in polypharmacology drug design.

Ranking Enzyme Structures in the PDB by Bound Ligand Similarity to Biological Substrates

There are numerous applications that use the structures of protein-ligand complexes from the PDB, such as 3D pharmacophore identification, virtual screening, and fragment-based drug design. The structures underlying these applications are potentially much more informative if they contain biologically relevant bound ligands, with high similarity to the cognate ligands. We present a study of ligand-enzyme complexes that compares the similarity of bound and cognate ligands, enabling the best matches to be identified. We calculate the molecular similarity scores using a method called PARITY (proportion of atoms residing in identical topology), which can conveniently be combined to give a similarity score for all cognate reactants or products in the reaction. Thus, we generate a rank-ordered list of related PDB structures, according to the biological similarity of the ligands bound in the structures.

•

We present PARITY, matching atoms in identical topology to gauge ligand similarity

•

Bound-cognate ligand similarity is a useful metric for ranking PDB structures

•

Only 26% of enzyme structures in the PDB have bound-cognate ligand similarity ≥0.7

•

We provide rank-ordered lists of PDBs with the most biologically relevant ligands

Understand protein functions by comparing the similarity of local structural environments

The three-dimensional structures of proteins play an essential role in regulating binding between proteins and their partners, offering a direct relationship between structures and functions of proteins. It is widely accepted that the function of a protein can be determined if its structure is similar to other proteins whose functions are known. However, it is also observed that proteins with similar global structures do not necessarily correspond to the same function, while proteins with very different folds can share similar functions. This indicates that function similarity is originated from the local structural information of proteins instead of their global shapes. We assume that proteins with similar local environments prefer binding to similar types of molecular targets. In order to testify this assumption, we designed a new structural indicator to define the similarity of local environment between residues in different proteins. This indicator was further used to calculate the probability that a given residue binds to a specific type of structural neighbors, including DNA, RNA, small molecules and proteins. After applying the method to a large-scale non-redundant database of proteins, we show that the positive signal of binding probability calculated from the local structural indicator is statistically meaningful. In summary, our studies suggested that the local environment of residues in a protein is a good indicator to recognize specific binding partners of the protein. The new method could be a potential addition to a suite of existing template-based approaches for protein function prediction.

Insights into cancer severity from biomolecular interaction mechanisms

To attain a deeper understanding of diseases like cancer, it is critical to couple genetics with biomolecular mechanisms. High-throughput sequencing has identified thousands of somatic mutations across dozens of cancers, and there is a pressing need to identify the few that are pathologically relevant. Here we use protein structure and interaction data to interrogate nonsynonymous somatic cancer mutations, identifying a set of 213 molecular interfaces (protein-protein, -small molecule or -nucleic acid) most often perturbed in cancer, highlighting several potentially novel cancer genes. Over half of these interfaces involve protein-small-molecule interactions highlighting their overall importance in cancer. We found distinct differences in the predominance of perturbed interfaces between cancers and histological subtypes and presence or absence of certain interfaces appears to correlate with cancer severity.

Prediction of Protein Pairs Sharing Common Active Ligands Using Protein Sequence, Structure, and Ligand Similarity

We benchmarked the ability of comparative computational approaches to correctly discriminate protein pairs sharing a common active ligand (positive protein pairs) from protein pairs with no common active ligands (negative protein pairs). Since the target and the off-targets of a drug share at least a common ligand, i.e., the drug itself, the prediction of positive protein pairs may help identify off-targets. We evaluated representative protein-centric and ligand-centric approaches, including (1) 2D and 3D ligand similarity, (2) several measures of protein sequence similarity in conjunction with different sequence sources (e.g., full protein sequence versus binding site residues), and (3) a newly described pocket shape similarity and alignment program called SiteHopper. While the sequence-based alignment of pocket residues achieved the best overall performance, SiteHopper outperformed sequence-based approaches for unrelated proteins with only 20-30% pocket residue identity. Analogously, among ligand-centric approaches, path-based fingerprints achieved the best overall performance, but ROCS-based ligand shape similarity outperformed path-based fingerprints for structurally dissimilar ligands (Tanimoto 25%-40%). A significant drop in recognition performance was observed for ligand-centric approaches when PDB ligands were used instead of ChEMBL ligands. Finally, we analyzed the relationship between pocket shape and ligand shape in our data set and found that similar ligands tend to bind to similar pockets while similar pockets may accept a range of different-shaped ligands.

Release of 50 new, drug-like compounds and their computational target predictions for open source anti-tubercular drug discovery

As a follow up to the antimycobacterial screening exercise and the release of GSK´s first Tres Cantos Antimycobacterial Set (TCAMS-TB), this paper presents the results of a second antitubercular screening effort of two hundred and fifty thousand compounds recently added to the GSK collection. The compounds were further prioritized based on not only antitubercular potency but also on physicochemical characteristics. The 50 most attractive compounds were then progressed for evaluation in three different predictive computational biology algorithms based on structural similarity or GSK historical biological assay data in order to determine their possible mechanisms of action. This effort has resulted in the identification of novel compounds and their hypothesized targets that will hopefully fuel future TB drug discovery and target validation programs alike.

A fast topological analysis algorithm for large-scale similarity evaluations of ligands and binding pockets

Motivation

With the rapid increase of the structural data of biomolecular complexes, novel structural analysis methods have to be devised with high-throughput capacity to handle immense data input and to construct massive networks at the minimal computational cost. Moreover, novel methods should be capable of handling a broad range of molecular structural sizes and chemical natures, cognisant of the conformational and electrostatic bases of molecular recognition, and sufficiently accurate to enable contextually relevant biological inferences.

Results

A novel molecular topology comparison method was developed and tested. The method was tested for both ligand and binding pocket similarity analyses and a PDB-wide ligand topological similarity map was computed.

Conclusion

The unprecedentedly wide scope of ligand definition and large-scale topological similarity mapping can provide very robust tools, of performance unmatched by the present alignment-based methods. The method remarkably shows potential for application for scaffold hopping purposes. It also opens new frontiers in the areas of ligand-mediated protein connectivity, ligand-based molecular phylogeny, target fishing, and off-target predictions.

Electronic supplementary material

The online version of this article (doi:10.1186/s13321-015-0091-5) contains supplementary material, which is available to authorized users.

Ligand-target prediction by structural network biology using nAnnoLyze

Target identification is essential for drug design, drug-drug interaction prediction, dosage adjustment and side effect anticipation. Specifically, the knowledge of structural details is essential for understanding the mode of action of a compound on a target protein. Here, we present nAnnoLyze, a method for target identification that relies on the hypothesis that structurally similar binding sites bind similar ligands. nAnnoLyze integrates structural information into a bipartite network of interactions and similarities to predict structurally detailed compound-protein interactions at proteome scale. The method was benchmarked on a dataset of 6,282 pairs of known interacting ligand-target pairs reaching a 0.96 of area under the Receiver Operating Characteristic curve (AUC) when using the drug names as an input feature for the classifier, and a 0.70 of AUC for “anonymous” compounds or compounds not present in the training set. nAnnoLyze resulted in higher accuracies than its predecessor, AnnoLyze. We applied the method to predict interactions for all the compounds in the DrugBank database with each human protein structure and provide examples of target identification for known drugs against human diseases. The accuracy and applicability of our method to any compound indicate that a comparative docking approach such as nAnnoLyze enables large-scale annotation and analysis of compound–protein interactions and thus may benefit drug development.

Description of the “mode-of-action” of a small chemical compound against a protein target is essential for the drug discovery process. Such description relies on three main steps: i) the identification of the target protein within the thousands of proteins in an organism, ii) the localization of the binding interaction site in the identified target protein, and iii) the molecular characterization of the compound’s binding mode in the binding site of the target protein. Here, we introduce a new computational method, called nAnnoLyze, which uses graph theory principles to relate compounds and target proteins based on comparative principles. nAnnoLyze aims at addressing two of the three previous steps, that is, target identification and binding site localization. Our results suggest that the nAnnoLyze accuracy and proteome-wide applicability enables the large-scale annotation and analysis of compound–protein interaction and thus may benefit drug development.

Polypharmacology modelling using proteochemometrics (PCM): recent methodological developments, applications to target families, and future prospects

Proteochemometric (PCM) modelling is a computational method to model the bioactivity of multiple ligands against multiple related protein targets simultaneously.

Mechismo: predicting the mechanistic impact of mutations and modifications on molecular interactions

Systematic interrogation of mutation or protein modification data is important to identify sites with functional consequences and to deduce global consequences from large data sets. Mechismo (mechismo.russellab.org) enables simultaneous consideration of thousands of 3D structures and biomolecular interactions to predict rapidly mechanistic consequences for mutations and modifications. As useful functional information often only comes from homologous proteins, we benchmarked the accuracy of predictions as a function of protein/structure sequence similarity, which permits the use of relatively weak sequence similarities with an appropriate confidence measure. For protein–protein, protein–nucleic acid and a subset of protein–chemical interactions, we also developed and benchmarked a measure of whether modifications are likely to enhance or diminish the interactions, which can assist the detection of modifications with specific effects. Analysis of high-throughput sequencing data shows that the approach can identify interesting differences between cancers, and application to proteomics data finds potential mechanistic insights for how post-translational modifications can alter biomolecular interactions.

PeptiSite: a structural database of peptide binding sites in 4D

We developed PeptiSite, a comprehensive and reliable database of biologically and structurally characterized peptide-binding sites, in which each site is represented by an ensemble of its complexes with protein, peptide and small molecule partners. The unique features of the database include: (1) the ensemble site representation that provides a fourth dimension to the otherwise three dimensional data, (2) comprehensive characterization of the binding site architecture that may consist of a multimeric protein assembly with cofactors and metal ions and (3) analysis of consensus interaction motifs within the ensembles and identification of conserved determinants of these interactions. Currently the database contains 585 proteins with 650 peptide-binding sites. http://peptisite.ucsd.edu/ link allows searching for the sites of interest and interactive visualization of the ensembles using the ActiveICM web-browser plugin. This structural database for protein-peptide interactions enables understanding of structural principles of these interactions and may assist the development of an efficient peptide docking benchmark.

Target prediction for an open access set of compounds active against Mycobacterium tuberculosis

Mycobacterium tuberculosis, the causative agent of tuberculosis (TB), infects an estimated two billion people worldwide and is the leading cause of mortality due to infectious disease. The development of new anti-TB therapeutics is required, because of the emergence of multi-drug resistance strains as well as co-infection with other pathogens, especially HIV. Recently, the pharmaceutical company GlaxoSmithKline published the results of a high-throughput screen (HTS) of their two million compound library for anti-mycobacterial phenotypes. The screen revealed 776 compounds with significant activity against the M. tuberculosis H37Rv strain, including a subset of 177 prioritized compounds with high potency and low in vitro cytotoxicity. The next major challenge is the identification of the target proteins. Here, we use a computational approach that integrates historical bioassay data, chemical properties and structural comparisons of selected compounds to propose their potential targets in M. tuberculosis. We predicted 139 target - compound links, providing a necessary basis for further studies to characterize the mode of action of these compounds. The results from our analysis, including the predicted structural models, are available to the wider scientific community in the open source mode, to encourage further development of novel TB therapeutics.

Mycobacterium tuberculosis is a major worldwide pathogen infecting millions individuals every year. Additionally, the number of antibiotic resistant strains has dramatically increased over the last decades. Trying to address this challenge, the pharmaceutical company GlaxoSmithKline has recently published the results of a large-scale high-throughput screen (HTS) that resulted in the release of 776 chemical compound structures active against tuberculosis. We have used this dataset of compounds as input to our computational approach that integrates historical bioassay data, chemical properties and structural comparisons. We propose 139 targets alongside their respective hit compounds and made them open to the wider scientific community. Our hope is that the availability of the experimental data from GSK and our computational analysis will encourage further research providing validated therapeutically targets against this devastating disease.

Drug Promiscuity in PDB: Protein Binding Site Similarity Is Key

Drug repositioning applies established drugs to new disease indications with increasing success. A pre-requisite for drug repurposing is drug promiscuity (polypharmacology) – a drug’s ability to bind to several targets. There is a long standing debate on the reasons for drug promiscuity. Based on large compound screens, hydrophobicity and molecular weight have been suggested as key reasons. However, the results are sometimes contradictory and leave space for further analysis. Protein structures offer a structural dimension to explain promiscuity: Can a drug bind multiple targets because the drug is flexible or because the targets are structurally similar or even share similar binding sites? We present a systematic study of drug promiscuity based on structural data of PDB target proteins with a set of 164 promiscuous drugs. We show that there is no correlation between the degree of promiscuity and ligand properties such as hydrophobicity or molecular weight but a weak correlation to conformational flexibility. However, we do find a correlation between promiscuity and structural similarity as well as binding site similarity of protein targets. In particular, 71% of the drugs have at least two targets with similar binding sites. In order to overcome issues in detection of remotely similar binding sites, we employed a score for binding site similarity: LigandRMSD measures the similarity of the aligned ligands and uncovers remote local similarities in proteins. It can be applied to arbitrary structural binding site alignments. Three representative examples, namely the anti-cancer drug methotrexate, the natural product quercetin and the anti-diabetic drug acarbose are discussed in detail. Our findings suggest that global structural and binding site similarity play a more important role to explain the observed drug promiscuity in the PDB than physicochemical drug properties like hydrophobicity or molecular weight. Additionally, we find ligand flexibility to have a minor influence.

Cheminformatics at the interface of medicinal chemistry and proteomics

Multiple factors have to be optimized in the course of a drug discovery project. Traditionally this includes potency on a single target, eventually specificity as well as the pharmacokinetic, physicochemical and the safety profile. Recently an additional dimension has been added by realizing that the therapeutic outcome of a drug is often determined not only by its activity on a single target but also by its activity profile across a variety of biological targets. To address the polypharmacology of drug candidates many compounds are tested on a set of targets or in phenotypic screens generating a tremendous amount of data. To extract useful information computational methods at the interface of proteomics and cheminformatics are indispensable. This review will focus on some recent developments in this field. This article is part of a Special Issue entitled: Computational Proteomics in the Post-Identification Era. Guest Editors: Martin Eisenacher and Christian Stephan.

Chemoisosterism in the proteome

The concept of chemoisosterism of protein environments is introduced as the complementary property to bioisosterism of chemical fragments. In the same way that two chemical fragments are considered bioisosteric if they can bind to the same protein environment, two protein environments will be considered chemoisosteric if they can interact with the same chemical fragment. The basis for the identification of chemoisosteric relationships among protein environments was the increasing amount of crystal structures available currently for protein-ligand complexes. It is shown that one can recover the right location and orientation of chemical fragments constituting the native ligand in a nuclear receptor structure by using only chemoisosteric environments present in enzyme structures. Examples of the potential applicability of chemoisosterism in fragment-based drug discovery are provided.

Pocketome: an encyclopedia of small-molecule binding sites in 4D

The importance of binding site plasticity in protein-ligand interactions is well-recognized, and so are the difficulties in predicting the nature and the degree of this plasticity by computational means. To assist in understanding the flexible protein-ligand interactions, we constructed the Pocketome, an encyclopedia of about one thousand experimentally solved conformational ensembles of druggable binding sites in proteins, grouped by location and consistent chain/cofactor composition. The multiplicity of pockets within the ensembles adds an extra, fourth dimension to the Pocketome entry data. Within each ensemble, the pockets were carefully classified by the degree of their pairwise similarity and compatibility with different ligands. The core of the Pocketome is derived regularly and automatically from the current releases of the Protein Data Bank and the Uniprot Knowledgebase; this core is complemented by entries built from manually provided seed ligand locations. The Pocketome website (www.pocketome.org) allows searching for the sites of interest, analysis of conformational clusters, important residues, binding compatibility matrices and interactive visualization of the ensembles using the ActiveICM web browser plugin. The Pocketome collection can be used to build multi-conformational docking and 3D activity models as well as to design cross-docking and virtual ligand screening benchmarks.

ProtChemSI: a network of protein-chemical structural interactions

Progress in structure determination methods means that the set of experimentally determined 3D structures of proteins in complex with small molecules is growing exponentially. ProtChemSI exploits and extends this useful set of structures by both collecting and annotating the existing data as well as providing models of potential complexes inferred by protein or chemical structure similarity. The database currently includes 7704 proteins from 1803 organisms, 11 324 chemical compounds and 202 289 complexes including 178 974 predicted. It is publicly available at http://pcidb.russelllab.org.

STITCH 3: zooming in on protein-chemical interactions

To facilitate the study of interactions between proteins and chemicals, we have created STITCH, an aggregated database of interactions connecting over 300 000 chemicals and 2.6 million proteins from 1133 organisms. Compared to the previous version, the number of chemicals with interactions and the number of high-confidence interactions both increase 4-fold. The database can be accessed interactively through a web interface, displaying interactions in an integrated network view. It is also available for computational studies through downloadable files and an API. As an extension in the current version, we offer the option to switch between two levels of detail, namely whether stereoisomers of a given compound are shown as a merged entity or as separate entities. Separate display of stereoisomers is necessary, for example, for carbohydrates and chiral drugs. Combining the isomers increases the coverage, as interaction databases and publications found through text mining will often refer to compounds without specifying the stereoisomer. The database is accessible at http://stitch.embl.de/.