Accelerated sampling of protein dynamics using BioEmu augmented molecular simulation

Abstract

We introduce a workflow that integrates BioEmu-generated conformational ensemble with physics-based molecular simulations and Markov State Models to sample Boltzmann-weighted conformational populations across biomolecules. Molecular simulations initiated from BioEmu ensemble capture active-to-inactive transitions in CDK2 and BRAF, two members of the serine-threonine kinase family, and elucidate how the disease-causing V600E mutation in BRAF drives population shifts among distinct metastable states relative to the wild type. Furthermore, we combined BioEmu ensemble with experimental cryo-EM data to construct all-atom conformational ensembles of biomolecules. In comparison to the AlphaFold2 reduced multiple sequence alignment (rMSA-AF2) approach, BioEmu-generated ensembles sample a broader conformational space for serine-threonine kinases but fail to capture conformational heterogeneity in several cases, including Glycine transporter 1 (GlyT1), a membrane transporter, and plasmepsin-II (PlmII), an aspartic protease. Systems where side-chain conformational heterogeneity governs protein dynamics such as cryptic pocket opening in PlmII or transitions between multiple metastable states in GlyT1; molecular simulations initiated from BioEmu generated ensemble do not capture the full spectrum of conformational heterogeneity. Overall, this study presents a straightforward framework for integrating generative AI based protein emulators with statistical physics to recover Boltzmann-weighted conformational ensembles at scale, while also highlighting critical limitations that necessitate careful, system-specific analysis when interpreting protein conformational landscapes.