Deep Local Analysis deconstructs protein - protein interfaces and accurately estimates binding affinity changes upon mutation

Abstract

The spectacular recent advances in protein and protein complex structure prediction hold promise for reconstructing interactomes at large scale and residue resolution. Beyond determining the 3D arrangement of interacting partners, modeling approaches should be able to unravel the impact of sequence variations on the strength of the association. In this work, we report on Deep Local Analysis (DLA), a novel and efficient deep learning framework that relies on a strikingly simple deconstruction of protein interfaces into small locally oriented residue-centered cubes and on 3D convolutions recognizing patterns within cubes. Merely based on the two cubes associated with the wild-type and the mutant residues, DLA accurately estimates the binding affinity change for the associated complexes. It achieves a Pearson correlation coefficient of 0.81 on more than 2 000 mutations, and its generalization capability to unseen complexes is higher than the state-of-the-art methods. We show that taking into account the evolutionary constraints on residues contributes to predictions. We also discuss the influence of conformational variability on performance. Beyond the predictive power on the effects of mutations, DLA is a general framework for transferring the knowledge gained from the available non-redundant set of complex protein structures to various tasks. For instance, given a single partially masked cube, it recovers the identity and physico-chemical class of the central residue. Given an ensemble of cubes representing an interface, it predicts the function of the complex. Source code and models are available at http://gitlab.lcqb.upmc.fr/DLA/DLA.git.

Benchmarking the accuracy of structure-based binding affinity predictors on Spike-ACE2 Deep Mutational Interaction Set

“…TopNetTree, a recent deep learning approach, includes physical pairwise interactions, Euclidean distances, and cavity structures within a topological framework. Notably, TopNetTree has been actively used in numerous SARS-CoV-2 studies[46,50,52–54,56]. In particular, Chen et al, trained TopNetTree on SARS-CoV-2 datasets to accurately predict changes in binding free energy for the S protein, ACE2, or antibodies induced by mutations[24]This tool was not available as a web server or a standalone tool.…”

“…During recent years, several AI methods have been proposed for predicting mutation-imposed interaction changes [21][22][23][46][47][48]. Among these tools, we concentrated on two, mmCSM-PPI [22] and TopNetTree [23,24], of which results on the RBD-ACE2 system were readily available.…”

Benchmarking the accuracy of structure‐based binding affinity predictors on Spike–ACE2 deep mutational interaction set

“…In recent years, several AI methods have been proposed for predicting mutation-imposed interaction changes. [21][22][23][46][47][48] Among these tools, we concentrated on two, mmCSM-PPI 22 and TopNetTree, 23,24 of which results on the RBD-ACE2 system were readily available. 24,[49][50][51][52][53][54][55][56] The machine learning approach, mmCSM-PPI, utilizes physicochemical and geometrical properties of protein structures within a graphbased structural framework to model the impact of mutations on the inter-residue interaction network.…”

Alignment-based protein mutational landscape prediction: doing more with less

“…Since then, we observed that the additional set of sequences had a limited impact on the performance (average ∆ρ = 0.012 on the dataset reported (Hopf et al, 2017)). Hence, in more recent studies (Tsuboyama et al, 2023;Mohseni Behbahani et al, 2023), we solely relied on an input alignment generated with the ProteinGym-MSA protocol. In the present work, for all calculations, we asked GEMME to exploit only a single input MSA generated by one of the four tested protocols and resources (see Additional file 1: Supplementary Methods for computational details).…”