InDeep: 3D fully convolutional neural networks to assist in silico drug design on protein-protein interactions.

MOTIVATION: Protein-protein interactions (PPIs) are key elements in numerous biological pathways and the subject of a growing number of drug discovery projects including against infectious diseases. Designing drugs on PPI targets remains a difficult task and requires extensive efforts to qualify a given interaction as an eligible target. To this end, besides the evident need to determine the role of PPIs in disease-associated pathways and their experimental characterization as therapeutics targets, prediction of their capacity to be bound by other protein partners or modulated by future drugs is of primary importance. RESULTS: We present InDeep, a tool for predicting functional binding sites within proteins that could either host protein epitopes or future drugs. Leveraging deep learning on a curated dataset of PPIs, this tool can proceed to enhanced functional binding site predictions either on experimental structures or along molecular dynamics trajectories. The benchmark of InDeep demonstrates that our tool outperforms state-of-the-art ligandable binding sites predictors when assessing PPI targets but also conventional targets. This offers new opportunities to assist drug design projects on PPIs by identifying pertinent binding pockets at or in the vicinity of PPI interfaces. AVAILABILITY AND IMPLEMENTATION: The tool is available on GitLab at https://gitlab.pasteur.fr/InDeep/InDeep. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

quicksom: Self-Organizing Maps on GPUs for clustering of molecular dynamics trajectories.

SUMMARY: We implemented the Self-Organizing Maps algorithm running efficiently on GPUs, and also provide several clustering methods of the resulting maps. We provide scripts and a use case to cluster macro-molecular conformations generated by molecular dynamics simulations. AVAILABILITY AND IMPLEMENTATION: The method is available on GitHub and distributed as a pip package.

Target Engagement and Binding Mode of an Antituberculosis Drug to Its Bacterial Target Deciphered in Whole Living Cells by NMR.

Detailed information on hit-target interaction is very valuable for drug discovery efforts and indispensable for rational hit to lead optimization. We developed a new approach combining NMR in whole-cells in-cell NMR) and docking to characterize hit-target interaction at the atomic level. By using in-cell NMR, we validated target engagement of the antituberculosis imidazopyridine amide (IPA) series with the subunit b of the cytochrome bc1:aa3, the major respiratory terminal oxidase in mycobacteria. The most advanced IPA called Q203 is currently in clinical trial. Using its derivative IPA317, we identified the atoms of the drug interacting with the cytochrome b in whole cells. NMR data and the self-organizing map algorithm were used to cluster a large set of drug-target complex models. The selected ensemble revealed IPA317 in a transient cavity of the cytochrome b, interacting directly with the residue T313, which is the site of spontaneous mutation conferring resistance to the IPA series. Our approach constitutes a pipeline to obtain atomic information on hit-target interactions in the cellular context.

Automated structure modeling of large protein assemblies using crosslinks as distance restraints.

Crosslinking mass spectrometry is increasingly used for structural characterization of multisubunit protein complexes. Chemical crosslinking captures conformational heterogeneity, which typically results in conflicting crosslinks that cannot be satisfied in a single model, making detailed modeling a challenging task. Here we introduce an automated modeling method dedicated to large protein assemblies ('XL-MOD' software is available at http://aria.pasteur.fr/supplementary-data/x-links) that (i) uses a form of spatial restraints that realistically reflects the distribution of experimentally observed crosslinked distances; (ii) automatically deals with ambiguous and/or conflicting crosslinks and identifies alternative conformations within a Bayesian framework; and (iii) allows subunit structures to be flexible during conformational sampling. We demonstrate our method by testing it on known structures and available crosslinking data. We also crosslinked and modeled the 17-subunit yeast RNA polymerase III at atomic resolution; the resulting model agrees remarkably well with recently published cryoelectron microscopy structures and provides additional insights into the polymerase structure.

Meet-U: Educating through research immersion.

We present a new educational initiative called Meet-U that aims to train students for collaborative work in computational biology and to bridge the gap between education and research. Meet-U mimics the setup of collaborative research projects and takes advantage of the most popular tools for collaborative work and of cloud computing. Students are grouped in teams of 4-5 people and have to realize a project from A to Z that answers a challenging question in biology. Meet-U promotes "coopetition," as the students collaborate within and across the teams and are also in competition with each other to develop the best final product. Meet-U fosters interactions between different actors of education and research through the organization of a meeting day, open to everyone, where the students present their work to a jury of researchers and jury members give research seminars. This very unique combination of education and research is strongly motivating for the students and provides a formidable opportunity for a scientific community to unite and increase its visibility. We report on our experience with Meet-U in two French universities with master's students in bioinformatics and modeling, with protein-protein docking as the subject of the course. Meet-U is easy to implement and can be straightforwardly transferred to other fields and/or universities. All the information and data are available at www.meet-u.org.

Improved large-scale prediction of growth inhibition patterns using the NCI60 cancer cell line panel.

MOTIVATION: Recent large-scale omics initiatives have catalogued the somatic alterations of cancer cell line panels along with their pharmacological response to hundreds of compounds. In this study, we have explored these data to advance computational approaches that enable more effective and targeted use of current and future anticancer therapeutics. RESULTS: We modelled the 50% growth inhibition bioassay end-point (GI50) of 17,142 compounds screened against 59 cancer cell lines from the NCI60 panel (941,831 data-points, matrix 93.08% complete) by integrating the chemical and biological (cell line) information. We determine that the protein, gene transcript and miRNA abundance provide the highest predictive signal when modelling the GI50 endpoint, which significantly outperformed the DNA copy-number variation or exome sequencing data (Tukey's Honestly Significant Difference, P <0.05). We demonstrate that, within the limits of the data, our approach exhibits the ability to both interpolate and extrapolate compound bioactivities to new cell lines and tissues and, although to a lesser extent, to dissimilar compounds. Moreover, our approach outperforms previous models generated on the GDSC dataset. Finally, we determine that in the cases investigated in more detail, the predicted drug-pathway associations and growth inhibition patterns are mostly consistent with the experimental data, which also suggests the possibility of identifying genomic markers of drug sensitivity for novel compounds on novel cell lines. CONTACT: terez@pasteur.fr; ab454@ac.cam.uk SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Neisseria meningitidis Type IV Pili Composed of Sequence Invariable Pilins Are Masked by Multisite Glycosylation.

The ability of pathogens to cause disease depends on their aptitude to escape the immune system. Type IV pili are extracellular filamentous virulence factors composed of pilin monomers and frequently expressed by bacterial pathogens. As such they are major targets for the host immune system. In the human pathogen Neisseria meningitidis, strains expressing class I pilins contain a genetic recombination system that promotes variation of the pilin sequence and is thought to aid immune escape. However, numerous hypervirulent clinical isolates express class II pilins that lack this property. This raises the question of how they evade immunity targeting type IV pili. As glycosylation is a possible source of antigenic variation it was investigated using top-down mass spectrometry to provide the highest molecular precision on the modified proteins. Unlike class I pilins that carry a single glycan, we found that class II pilins display up to 5 glycosylation sites per monomer on the pilus surface. Swapping of pilin class and genetic background shows that the pilin primary structure determines multisite glycosylation while the genetic background determines the nature of the glycans. Absence of glycosylation in class II pilins affects pilus biogenesis or enhances pilus-dependent aggregation in a strain specific fashion highlighting the extensive functional impact of multisite glycosylation. Finally, molecular modeling shows that glycans cover the surface of class II pilins and strongly decrease antibody access to the polypeptide chain. This strongly supports a model where strains expressing class II pilins evade the immune system by changing their sugar structure rather than pilin primary structure. Overall these results show that sequence invariable class II pilins are cloaked in glycans with extensive functional and immunological consequences.

An automatic tool to analyze and cluster macromolecular conformations based on self-organizing maps.

MOTIVATION: Sampling the conformational space of biological macromolecules generates large sets of data with considerable complexity. Data-mining techniques, such as clustering, can extract meaningful information. Among them, the self-organizing maps (SOMs) algorithm has shown great promise; in particular since its computation time rises only linearly with the size of the data set. Whereas SOMs are generally used with few neurons, we investigate here their behavior with large numbers of neurons. RESULTS: We present here a python library implementing the full SOM analysis workflow. Large SOMs can readily be applied on heavy data sets. Coupled with visualization tools they have very interesting properties. Descriptors for each conformation of a trajectory are calculated and mapped onto a 3D landscape, the U-matrix, reporting the distance between neighboring neurons. To delineate clusters, we developed the flooding algorithm, which hierarchically identifies local basins of the U-matrix from the global minimum to the maximum. AVAILABILITY AND IMPLEMENTATION: The python implementation of the SOM library is freely available on github: https://github.com/bougui505/SOM. CONTACT: michael.nilges@pasteur.fr or guillaume.bouvier@pasteur.fr SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Building Graphs To Describe Dynamics, Kinetics, and Energetics in the d-ALa:d-Lac Ligase VanA.

The d-Ala:d-Lac ligase, VanA, plays a critical role in the resistance of vancomycin. Indeed, it is involved in the synthesis of a peptidoglycan precursor, to which vancomycin cannot bind. The reaction catalyzed by VanA requires the opening of the so-called "ω-loop", so that the substrates can enter the active site. Here, the conformational landscape of VanA is explored by an enhanced sampling approach: the temperature-accelerated molecular dynamics (TAMD). Analysis of the molecular dynamics (MD) and TAMD trajectories recorded on VanA permits a graphical description of the structural and kinetics aspects of the conformational space of VanA, where the internal mobility and various opening modes of the ω-loop play a major role. The other important feature is the correlation of the ω-loop motion with the movements of the opposite domain, defined as containing the residues A149-Q208. Conformational and kinetic clusters have been determined and a path describing the ω-loop opening was extracted from these clusters. The determination of this opening path, as well as the relative importance of hydrogen bonds along the path, permit one to propose some key residue interactions for the kinetics of the ω-loop opening.

An algorithm to enumerate all possible protein conformations verifying a set of distance constraints.

BACKGROUND: The determination of protein structures satisfying distance constraints is an important problem in structural biology. Whereas the most common method currently employed is simulated annealing, there have been other methods previously proposed in the literature. Most of them, however, are designed to find one solution only. RESULTS: In order to explore exhaustively the feasible conformational space, we propose here an interval Branch-and-Prune algorithm (iBP) to solve the Distance Geometry Problem (DGP) associated to protein structure determination. This algorithm is based on a discretization of the problem obtained by recursively constructing a search space having the structure of a tree, and by verifying whether the generated atomic positions are feasible or not by making use of pruning devices. The pruning devices used here are directly related to features of protein conformations. CONCLUSIONS: We described the new algorithm iBP to generate protein conformations satisfying distance constraints, that would potentially allows a systematic exploration of the conformational space. The algorithm iBP has been applied on three α-helical peptides.

Identification of binding sites and favorable ligand binding moieties by virtual screening and self-organizing map analysis.

BACKGROUND: Identifying druggable cavities on a protein surface is a crucial step in structure based drug design. The cavities have to present suitable size and shape, as well as appropriate chemical complementarity with ligands. RESULTS: We present a novel cavity prediction method that analyzes results of virtual screening of specific ligands or fragment libraries by means of Self-Organizing Maps. We demonstrate the method with two thoroughly studied proteins where it successfully identified their active sites (AS) and relevant secondary binding sites (BS). Moreover, known active ligands mapped the AS better than inactive ones. Interestingly, docking a naive fragment library brought even more insight. We then systematically applied the method to the 102 targets from the DUD-E database, where it showed a 90% identification rate of the AS among the first three consensual clusters of the SOM, and in 82% of the cases as the first one. Further analysis by chemical decomposition of the fragments improved BS prediction. Chemical substructures that are representative of the active ligands preferentially mapped in the AS. CONCLUSION: The new approach provides valuable information both on relevant BSs and on chemical features promoting bioactivity.

Stabilization of the integrase-DNA complex by Mg2+ ions and prediction of key residues for binding HIV-1 integrase inhibitors.

The HIV-1 integrase is an attractive target for the therapeutics development against AIDS, as no host homologue of this protein has been identified. The integrase strand transfer inhibitors (INSTIs), including raltegravir, specifically target the second catalytic step of the integration process by binding to the DDE motif of the catalytic site and coordinating Mg(2+) ions. Recent X-ray crystallographic structures of the integrase/DNA complex from prototype foamy virus allowed to investigate the role of the different partners (integrase, DNA, Mg(2+) ions, raltegravir) in the complex stability using molecular dynamics (MD) simulations. The presence of Mg(2+) ions is found to be essential for the stability, whereas the simultaneous presence of raltegravir and Mg(2+) ions has a destabilizing influence. A homology model of HIV-1 integrase was built on the basis of the X-ray crystallographic information, and protein marker residues for the ligand binding were detected by clustering the docking poses of known HIV-1 integrase inhibitors on the model. Interestingly, we had already identified some of these residues to be involved in HIV-1 resistance mutations and in the stabilization of the catalytic site during the MD simulations. Classification of protein conformations along MD simulations, as well as of ligand docking poses, was performed by using an original learning method, based on self-organizing maps. This allows us to perform a more in-depth investigation of the free-energy basins populated by the complex in MD simulations on the one hand, and a straightforward classification of ligands according to their binding residues on the other hand.

Functional motions modulating VanA ligand binding unraveled by self-organizing maps.

The VanA D-Ala:D-Lac ligase is a key enzyme in the emergence of high level resistance to vancomycin in Enterococcus species and methicillin-resistant Staphylococcus aureus. It catalyzes the formation of D-Ala-D-Lac instead of the vancomycin target, D-Ala-D-Ala, leading to the production of modified, low vancomycin binding affinity peptidoglycan precursors. Therefore, VanA appears as an attractive target for the design of new antibacterials to overcome resistance. The catalytic site of VanA is delimited by three domains and closed by an ω-loop upon enzymatic reaction. The aim of the present work was (i) to investigate the conformational transition of VanA associated with the opening of its ω-loop and of a part of its central domain and (ii) to relate this transition with the substrate or product binding propensities. Molecular dynamics trajectories of the VanA ligase of Enterococcus faecium with or without a disulfide bridge distant from the catalytic site revealed differences in the catalytic site conformations with a slight opening. Conformations were clustered with an original machine learning method, based on self-organizing maps (SOM), which revealed four distinct conformational basins. Several ligands related to substrates, intermediates, or products were docked to SOM representative conformations with the DOCK 6.5 program. Classification of ligand docking poses, also performed with SOM, clearly distinguished ligand functional classes: substrates, reaction intermediates, and product. This result illustrates the acuity of the SOM classification and supports the quality of the DOCK program poses. The protein-ligand interaction features for the different classes of poses will guide the search and design of novel inhibitors.

AlignScape, displaying sequence similarity using self-organizing maps.

The current richness of sequence data needs efficient methodologies to display and analyze the complexity of the information in a compact and readable manner. Traditionally, phylogenetic trees and sequence similarity networks have been used to display and analyze sequences of protein families. These methods aim to shed light on key computational biology problems such as sequence classification and functional inference. Here, we present a new methodology, AlignScape, based on self-organizing maps. AlignScape is applied to three large families of proteins: the kinases and GPCRs from human, and bacterial T6SS proteins. AlignScape provides a map of the similarity landscape and a tree representation of multiple sequence alignments These representations are useful to display, cluster, and classify sequences as well as identify functional trends. The efficient GPU implementation of AlignScape allows the analysis of large MSAs in a few minutes. Furthermore, we show how the AlignScape analysis of proteins belonging to the T6SS complex can be used to predict coevolving partners.

Mechanism of action of phthalazinone derivatives against rabies virus.

Rabies, a viral zoonosis, is responsible for almost 59,000 deaths each year, despite the existence of an effective post-exposure prophylaxis. Indeed, rabies causes acute encephalomyelitis, with a case-fatality rate of 100 % after the onset of neurological clinical signs. Therefore, the development of therapies to inhibit the rabies virus (RABV) is crucial. Here, we identified, from a 30,000 compound library screening, phthalazinone derivative compounds as potent inhibitors of RABV infection and more broadly of Lyssavirus and even Mononegavirales infections. Combining in vitro experiments, structural modelling, in silico docking and in vivo assays, we demonstrated that phthalazinone derivatives display a strong inhibition of lyssaviruses infection by acting directly on the replication complex of the virus, and with noticeable effects in delaying the onset of the clinical signs in our mouse model.

A convective replica-exchange method for sampling new energy basins.

Replica-exchange is a powerful simulation method for sampling the basins of a rugged energy landscape. The replica-exchange method's sampling is efficient because it allows replicas to perform round trips in temperature space, thereby visiting both low and high temperatures in the same simulation. However, replicas have a diffusive walk in temperature space, and the round trip rate decreases significantly with the system size. These drawbacks make convergence of the simulation even more difficult than it already is when bigger systems are tackled. Here, we present a simple modification of the exchange method. In this method, one of the replicas steadily raises or lowers its temperature. We tested the convective replica-exchange method on three systems of varying complexity: the alanine dipeptide in implicit solvent, the GB1 ß-hairpin in explicit solvent and the Aß(25-35) homotrimer in a coarse grained representation. For the highly frustrated Aß(25-35) homotrimer, the proposed "convective" replica-exchange method is twice as fast as the standard method. It discovered 24 out of 27 free-energy basins in less than 500 ns. It also prevented the formation of groups of replicas that usually form on either side of an exchange bottleneck, leading to a more efficient sampling of new energy basins than in the standard method.

Building Protein Atomic Models from Cryo-EM Density Maps and Residue Co-Evolution.

Electron cryo-microscopy (cryo-EM) has emerged as a powerful method by which to obtain three-dimensional (3D) structures of macromolecular complexes at atomic or near-atomic resolution. However, de novo building of atomic models from near-atomic resolution (3-5 Å) cryo-EM density maps is a challenging task, in particular because poorly resolved side-chain densities hamper sequence assignment by automatic procedures at a lower resolution. Furthermore, segmentation of EM density maps into individual subunits remains a difficult problem when the structure of the subunits is not known, or when significant conformational rearrangement occurs between the isolated and associated form of the subunits. To tackle these issues, we have developed a graph-based method to thread most of the C-α trace of the protein backbone into the EM density map. The EM density is described as a weighted graph such that the resulting minimum spanning tree encompasses the high-density regions of the map. A pruning algorithm cleans the tree and finds the most probable positions of the C-α atoms, by using side-chain density when available, as a collection of C-α trace fragments. By complementing experimental EM maps with contact predictions from sequence co-evolutionary information, we demonstrate that this approach can correctly segment EM maps into individual subunits and assign amino acid sequences to backbone traces to generate atomic models.

Deciphering the Molecular Interaction Between the Adhesion G Protein-Coupled Receptor ADGRV1 and its PDZ-Containing Regulator PDZD7.

Hearing relies on the transduction of sound-evoked vibrations into electrical signals, occurring in the stereocilia bundle of inner ear hair cells. The G protein-coupled receptor (GPCR) ADGRV1 and the multi-PDZ protein PDZD7 play a critical role in the formation and function of stereocilia through their scaffolding and signaling properties. During hair cell development, the GPCR activity of ADGRV1 is specifically inhibited by PDZD7 through an unknown mechanism. Here, we describe the key interactions mediated by the two N-terminal PDZ domains of PDZD7 and the cytoplasmic domain of ADGRV1. Both PDZ domains can bind to the C-terminal PDZ binding motif (PBM) of ADGRV1 with the critical contribution of atypical C-terminal ß extensions. The two PDZ domains form a supramodule in solution, stabilized upon PBM binding. Interestingly, we showed that the stability and binding properties of the PDZ tandem are affected by two deafness-causing mutations located in the binding grooves of PDZD7 PDZ domains.

Automatic clustering of docking poses in virtual screening process using self-organizing map.

MOTIVATION: Scoring functions provided by the docking software are still a major limiting factor in virtual screening (VS) process to classify compounds. Score analysis of the docking is not able to find out all active compounds. This is due to a bad estimation of the ligand binding energies. Making the assumption that active compounds should have specific contacts with their target to display activity, it would be possible to discriminate active compounds from inactive ones with careful analysis of interatomic contacts between the molecule and the target. However, compounds clustering is very tedious due to the large number of contacts extracted from the different conformations proposed by docking experiments. RESULTS: Structural analysis of docked structures is processed in three steps: (i) a Kohonen self-organizing map (SOM) training phase using drug-protein contact descriptors followed by (ii) an unsupervised cluster analysis and (iii) a Newick file generation for results visualization as a tree. The docking poses are then analysed and classified quickly and automatically by AuPosSOM (Automatic analysis of Poses using SOM). AuPosSOM can be integrated into strategies for VS currently employed. We demonstrate that it is possible to discriminate active compounds from inactive ones using only mean protein contacts' footprints calculation from the multiple conformations given by the docking software. Chemical structure of the compound and key binding residues information are not necessary to find out active molecules. Thus, contact-activity relationship can be employed as a new VS process. AVAILABILITY: AuPosSOM is available at http://www.aupossom.com.

CC17 group B Streptococcus exploits integrins for neonatal meningitis development.

Group B Streptococcus (GBS) is the major cause of human neonatal infections. A single clone, designated CC17-GBS, accounts for more than 80% of meningitis cases, the most severe form of the infection. However, the events allowing blood-borne GBS to penetrate the brain remain largely elusive. In this study, we identified the host transmembrane receptors α5ß1 and αvß3 integrins as the ligands of Srr2, a major CC17-GBS-specific adhesin. Two motifs located in the binding region of Srr2 were responsible for the interaction between CC17-GBS and these integrins. We demonstrated in a blood-brain-barrier cellular model that both integrins contributed to the adhesion and internalization of CC17-GBS. Strikingly, both integrins were overexpressed during the postnatal period in the brain vessels of the blood-brain barrier and blood-cerebrospinal fluid barrier and contributed to juvenile susceptibility to CC17 meningitis. Finally, blocking these integrins decreased the ability of CC17-GBS to cross into the CNS of juvenile mice in an in vivo model of meningitis. Our study demonstrated that CC17-GBS exploits integrins in order to cross the brain vessels, leading to meningitis. Importantly, it provides host molecular insights into neonate's susceptibility to CC17-GBS meningitis, thereby opening new perspectives for therapeutic and prevention strategies of GBS-elicited meningitis.