Learning residue level protein dynamics with multiscale Gaussians

ICLR 2026 Conference Submission19931 Authors

Keywords: protein dynamics, flexibility, ensembles

TL;DR: Lightweight SE(3) invariant model for learning protein dynamics Gaussians and fast ensemble generation

Abstract:

Many methods have been developed to predict static protein structures, however understanding the \textit{dynamics} of protein structure is essential for elucidating biological function. While molecular dynamics (MD) simulations remain the \textit{in silico} gold standard, its high computational cost limits scalability. We present \textsc{DynaProt}, a lightweight, SE(3)-invariant framework that predicts rich descriptors of protein dynamics directly from static structures. By casting the problem through the lens of multivariate Gaussians, \textsc{DynaProt} estimates dynamics at two complementary scales: (1) per-residue marginal anisotropy as
covariance matrices capturing local flexibility, and (2) joint scalar covariances encoding pairwise dynamic coupling across residues. From these dynamics outputs, \textsc{DynaProt} achieves high accuracy in predicting residue-level flexibility (RMSF) and, remarkably, enables reasonable reconstruction of the full covariance matrix for fast ensemble generation. Notably, it does so using orders of magnitude fewer parameters than prior methods. Our results highlight the potential of direct protein dynamics prediction as a scalable alternative to existing methods.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 19931