Abstract
Nearly half of the proteome is intrinsically disordered, consisting of proteins or motifs that lack a stable secondary structure. Modeling these regions remains challenging, even with tremendous advances in generative models for protein structure prediction. Sampling the conformational ensemble of the intrinsically disordered proteome is out of reach for atomistic molecular dynamics due to the deeply metastable probability distributions over conformational space. In this talk, I will show how generative models with physical latent space dynamics provide a generalizable strategy for robustly sampling conformational ensembles. Furthermore, I will detail how to construct this sampling scheme to ensure asymptotic convergence of the samples to a Boltzmann distribution.