Guillaume Lamoureux
Chemistry
Rutgers University
Camden, New Jersey

Monday, October 17, 2022
3:00 pm (CT)
Hybrid webinar recording

Abstract

Protein-protein interactions (PPIs) are essential for biological function but remain very difficult to predict computationally. Coevolution signals between two interacting proteins are much weaker than within a single protein and can be detected reliably only if a large number of orthologous protein pairs are available. PPIs are also heavily dependent on biological context and are sensitive to perturbations such as genetic variations or disease mutations. In this talk, I will present our contributions to the development of unified 'sequence-to-structure-to-function' models based on deep neural networks. These models aim at predicting how proteins assemble and interact with one another using molecular representations learned from high-throughput PPI data. We have demonstrated the concept on a simplified version of the protein docking problem and I will discuss our efforts at applying it to more challenging prediction tasks.

https://youtu.be/A9-MfeuDLtc