Top Cited Papers
Open Access
Abstract
The most popular and universally predictive protein simulation models employ all-atom molecular dynamics, but they come at extreme computational cost. The development of a universal, computationally efficient coarse-grained (CG) model with similar prediction performance has been a long-standing challenge. By combining recent deep-learning methods with a large and diverse training set of all-atom protein simulations, we here develop a bottom–up CG force field with chemical transferability, which can be used for extrapolative molecular dynamics on new sequences not used during model parameterization. We demonstrate that the model successfully predicts metastable states of folded, unfolded and intermediate structures, the fluctuations of intrinsically disordered proteins and relative folding free energies of protein mutants, while being several orders of magnitude faster than an all-atom model. This showcases the feasibility of a universal and computationally efficient machine-learned CG model for proteins.
Funding Information
  • Bundesministerium für Forschung, Technologie und Raumfahrt (FAIME 01IS24076, FAIME 01IS24076, FAIME 01IS24076, FAIME 01IS24076)
  • Deutsche Forschungsgemeinschaft (SFB/TRR 186, Project A12, SFB 1114, Projects B03, B08, and A04, RTG 2433, Project Q05, SFB 1078, Project C7, RTG 2433, SFB 1114, SFB 1114, SFB 1114, SFB 1078, RTG 2433, Berlin Mathematics center MATH+ (AA1-6 and EF1-2), 825/2, 825/3, 825/4)
  • Einstein Stiftung Berlin (Project 0420815101, Project 0420815101, Project 0420815101, Project 0420815101, Project 0420815101, Project 0420815101, Project 0420815101)
  • National Science Foundation (CHE-1900374, PHY-2019745)
  • Max-Planck-Gesellschaft (IMPRS–BAC, IMPRS-BAC, IMPRS-BAC, IMPRS-BAC)
  • European Commission (ERC CoG 772230, ERC CoG 772230, ERC CoG 772230)

This publication has 63 references indexed in Scilit: