af2rave protocol generates reaction coordinates transferable across select kinases

Learning a reaction coordinate (RC) for efficient sampling of proteins often requires much domain knowledge, and as such learned coordinates are often system-specific. The af2rave protocol learns an RC to sample and rank protein conformations by combining AlphaFold2-type models with physics-based sampling techniques. Molecular simulations from the AlphaFold2 generated conformations can be used to learn relevant slow degrees of freedom, or RCs. The learned RC can then be biased against in an enhanced sampling simulation. With af2rave, an open source python package developed by us, we learned an RC for the DDR1 kinase and performed metadynamics simulations to explore the activation loop and Asp-Phe-Gly (DFG) motif movement. Further, we demonstrated the transferability of this RC by applying this protocol to the Abl1 and src kinases.