Machine Learning in Montpellier, Theory & Practice – Vincent Mallet (Ecole des Mines de Paris)
Learning on the 3D structure of biomolecules resulted in several breakthroughs, for instance the de-novo design of potent binders or functional luciferases, eventually leading to a Nobel Prize. Most of this work relies on modeling the structure of protein as graphs, with well-established tools. In my talk I will start by introducing the topic and reviewing recent methods. I will then present our recent results on the coordinated use of graphs, sequence and surface representations to model protein structure with machine learning. We show that despite disappointing results in isolation, surface methods used in addition to others display a synergistic effect. I will also discuss ongoing work to enhance this method, as well as apply it to homologs mining and binder design. About the presenter: Vincent Mallet is a researcher working on modeling the structure of biomolecules with machine learning, with applications in structural biology and drug design. His two main lines of work are surface-based representations for proteins and graph-based representations for RNA. He is part of the CBIO team, in Mines Paris PSL and Institut Curie. Previously, he was a postdoc with Maks Ovsjanikov working on geometric deep learning. He did his PhD with Jean-Philippe Vert and Michael Nilges on equivariant methods and drug design.
