Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning

Gainza, Pablo; Sverrisson, Freyr; Monti, Federico; Rodolà, Emanuele; Boscaini, Davide; Bronstein, Michael M.; Correia, Bruno E.

doi:10.1038/s41592-019-0666-6

Cited by 477 publications

(555 citation statements)

References 57 publications

Supporting

Mentioning

551

Contrasting

Order By: Relevance

“…Recent machine learning approaches, such as deep learning approaches 54,55 , can increase the coverage of the structurally-resolved human interactome for future studies.…”

Section: Discussionmentioning

confidence: 99%

Comprehensive characterization of protein-protein interaction network perturbations by human disease mutations

Cheng

Zhao

Wang

et al. 2020

Preprint

View full text Add to dashboard Cite

Technological and computational advances in genomics and interactomics have made it possible to identify rapidly how disease mutations perturb interaction networks within human cells. In this study, we investigate at large-scale the effects of network perturbations caused by disease mutations within the human three-dimensional (3D), structurally-resolved macromolecular interactome. We show that disease-associated germline mutations are significantly enriched in sequences encoding protein-protein interfaces compared to mutations identified in healthy subjects from the 1000 Genomes and ExAC projects; these interface mutations correspond to protein-protein interaction (PPI)-perturbing alleles including p.Ser127Arg in PCSK9 at the PCSK9-LDLR interface. In addition, somatic missense mutations are significantly enriched in PPI interfaces compared to non-interfaces in 10,861 human exomes across 33 cancer subtypes/types from The Cancer Genome Atlas. Using a binomial statistical model, we computationally identified 470 PPIs harboring a statistically significant excess number of missense mutations at protein-protein interfaces (termed putative oncoPPIs) in pan-cancer analysis. We demonstrate that the oncoPPIs, including histone H4 complex in individual cancer types, are highly correlated with patient survival and drug resistance/sensitivity in human cancer cell lines and patient-derived xenografts. We experimentally validate the network effects of 13 oncoPPIs using a systematic binary interaction assay. We further showed that ALOX5 p.Met146Lys at the ALOX5-MAD1L1 interface and RXRA p.Ser427Phe at the RXRA-PPARG interface promote significant tumor cell growth using cell line-based functional assays, providing a functional proof-of-concept. In summary, if broadly applied, this human 3D interactome network analysis offers a powerful tool for prioritizing alleles with mutations altering PPIs that may contribute to the pathobiology of human diseases, and may offer disease-specific targets for genotype-informed therapeutic discovery.

show abstract

“…Recent machine learning approaches, such as deep learning approaches 54,55 , can increase the coverage of the structurally-resolved human interactome for future studies.…”

Section: Discussionmentioning

confidence: 99%

Comprehensive characterization of protein-protein interaction network perturbations by human disease mutations

Cheng

Zhao

Wang

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Likewise, ComBind can be used with any pairwise pose similarity metric or combination thereof. ComBind's performance could potentially be improved by using more fine-grained interaction descriptors (41,42) or by using similarity metrics based on field-based methods developed for virtual screening (28,43).…”

Section: Extensibility and Future Workmentioning

confidence: 99%

Leveraging non-structural data to predict structures of protein–ligand complexes

et al. 2020

Preprint

View full text Add to dashboard Cite

Structure-based drug design depends on the ability to predict the three-dimensional structure of ligands bound to their targets, as does understanding the molecular mechanisms of many essential biological processes. Dozens of computational docking methods have been developed to address this binding pose prediction problem, but they frequently produce inaccurate results. Here we present a method that substantially improves the accuracy of binding pose prediction by exploiting a widely available source of non-structural information: a list of other ligands that bind the same target. Our method, ComBind, quantifies and leverages the chemist's intuition that even very different ligands tend to form similar interactions with a target protein. We demonstrate that ComBind consistently increases pose prediction accuracy across all major families of drug targets. We also illustrate its use by predicting previously unknown binding poses of antipsychotics and validating these results experimentally.

show abstract

“…Recently, a number of tools have been proposed for protein-protein interface comparisons. MaSIF compares molecular surfaces of protein-protein interfaces using geometric deep learning [21]. Whilst this technique has shown success using bound structures to search for similar binders, the performance did not hold on the more challenging unbound set.…”

Section: Introductionmentioning

confidence: 99%

Ab-Ligity: Identifying sequence-dissimilar antibodies that bind to the same epitope

Wong

Robinson

Bujotzek

et al. 2020

Preprint

View full text Add to dashboard Cite

Motivation: Solving the structure of an antibody-antigen complex gives atomic level information of the interactions between an antibody and its antigen, but such structures are expensive and hard to obtain. Alternative experimental sources include epitope mapping and binning experiments which can be used as a surrogate to identify key interacting residues. However, their resolution is usually not sufficient to identify if two antibodies have identical interactions. Computational approaches to this problem have so far been based on the premise that antibodies with similar sequences behave similarly. Such approaches will fail to identify sequence-distant antibodies that target the same epitope. Results: We present Ab-Ligity, a structure-based similarity measure tailored to antibodyantigen interfaces. Using predicted paratopes on model antibody structures, we assessed our ability to identify those antibodies that target highly similar epitopes. Most antibodies adopting similar binding modes can be identified from sequence similarity alone, using methods such as clonotyping. In the challenging subset where the antibody sequences differ significantly, Ab-Ligity is still able to predict antibodies that would bind to highly similar epitopes (area under the precision-recall curve of 0.86). We compared Ab-Ligity's performance to an existing tool InterComp, and showed improved performance alongside a significant speed-up. These results suggest that Ab-Ligity will allow the identification of diverse (sequence-dissimilar) antibodies that bind to the same epitopes from large datasets such as immune repertoires. Availability: http://opig.stats.ox.ac.uk/resources Contact: deane@stats.ox.ac.uk Supplementary information: Supplementary Materials are available at Biorxiv.

show abstract

Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning

Cited by 477 publications

References 57 publications

Comprehensive characterization of protein-protein interaction network perturbations by human disease mutations

Comprehensive characterization of protein-protein interaction network perturbations by human disease mutations

Leveraging non-structural data to predict structures of protein–ligand complexes

Ab-Ligity: Identifying sequence-dissimilar antibodies that bind to the same epitope

Contact Info

Product

Resources

About