“Geometric and Topological Machine Learning for Drug Discovery and Pattern Formation”
Learning the intrinsic geometry and topology of biological data is crucial for uncovering new insights in biomedicine. This talk will explore advanced geometric and topological machine learning methods tailored for drug design, understanding protein conformational landscapes, and cell signaling dynamics. I will begin by presenting the learnable geometric scattering transform, which underpins both GRASSY and ProtSCAPE. GRASSY is a generative model for in silico drug design and lead optimization, and ProtSCAPE leverages structured latent representations learned from molecular dynamics trajectories to elucidate protein dynamics. Next, I will introduce GSTH - a novel approach that integrates geometric scattering with manifold learning and persistent homology - to capture and analyze the topological signatures intrinsic to the spatiotemporal dynamics of cell signaling.
Host: Naomi Gluck