Geometric Deep Learning for Biomolecules

Abstract

Recent advancements in machine learning offer a promising pathway to deeper insights into biological phenomena. This manuscript explores the integration of geometric deep learning techniques to model biological structures. By embedding inductive biases based on geometry and physical laws, we aim to enhance our understanding and predictive capabilities in biomolecular systems. We present methods using equivariant neural networks for geometrical protein representation learning, molecular representation learning for electron density prediction, and scalable molecular dynamics simulations using stochastic interpolants.