FlowBack: A Generalized Flow-Matching Approach for Biomolecular Backmapping

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 7, 2025

Coarse-grained models have become ubiquitous in biomolecular modeling tasks aimed at studying slow dynamical processes such as protein folding and DNA hybridization. These can considerably accelerate sampling but it remains challenging to accurately efficiently restore all-atom detail the coarse-grained trajectory, which be vital for detailed understanding of molecular mechanisms calculation observables contingent on coordinates. In this work, we introduce FlowBack a deep generative model employing flow-matching objective map samples from prior distribution an data distribution. We construct our agnostic type. A protein-specific trained ∼65k structures Protein Data Bank achieves state-of-the-art performance structural metrics compared previous rules-based approaches applications static PDB structures, simulations fast-folding proteins, trajectories generated by machine-learned force field. DNA–protein ∼1.5k complexes excellent reconstruction capabilities well out-of-distribution complexation. offers accurate, efficient, easy-to-use tool recover with higher robustness fewer steric clashes than approaches. make freely available community open source Python package.

Language: Английский

Emerging Artificial Intelligence Methodologies in Computational Biology

Journal of Molecular Biology, Journal Year: 2025, Volume and Issue: unknown, P. 169002 - 169002

Published: Feb. 1, 2025

Language: Английский