Scoring and Predicting Protein Interactions and Conformations based on Evolutionary Signals – SPPICES

AlphaFold2 has revolutionised the field of structural bioinformatics and biology in general, accurately predicting protein structures from sequence information. However, it fails to unveil the full complexities of protein movements and interactions that are key to understanding the molecular mechanisms of protein function. SPPICES aims to fill that gap by using evolutionary information to help AlphaFold2 predict alternative conformations and enhance its sampling of dynamical protein-protein interfaces along the whole order-disorder continuum. SPPICES primary hypothesis is that evolution shapes biologically relevant protein conformations so that we can extract evolutionary signals to improve our predictive models. Building on AlphaFold2’s strengths and integrating state-of-the-art deep learning approaches, particularly graph-based techniques, three interconnected work packages (WPs) will be executed. Firstly, WP1 will focus on predicting conformations of mono- and multi-domain proteins using AlphaFold2 by exploiting structural templates and multiple sequence alignment editions. Simultaneously, WP2 will delve into predicting conformational changes involving order-disorder transitions and selecting functional conformations from the structural ensemble of intrinsically disordered proteins using novel evolutionary metrics. Lastly, WP3 will pivot towards modelling and scoring protein-protein interactions, integrating outputs from the previous WPs to model dynamical and fuzzy interfaces. By bridging deep learning methods and evolutionary biology, SPPICES seeks to improve our comprehension of protein dynamics, interactions, and functions, ultimately benefiting the broader scientific community.