PathGennie: Rapid Generation of Rare Event Pathways via Direction-Guided Adaptive Sampling Using Ultrashort Monitored Trajectories

Generating mechanistic pathways of rare events like ligand unbinding or protein folding is a central challenge in molecular dynamics simulation of complex molecular processes. Here, we present PathGennie, a direction-guided adaptive sampling framework that rapidly generates transition pathways without external biasing forces or thermal perturbation. The method launches swarms of ultrashort (a few femtoseconds long), unbiased trajectories and selectively propagates only those that exhibit progress toward a defined goal in a possibly high-dimensional collective variable space, thus circumventing long waiting times while preserving the true dynamics. We demonstrate the broad applicability and efficiency of PathGennie by identifying multiple competing unbinding pathways for benzene in T4 lysozyme and imatinib from Abl kinase, as well as capturing the folding and unfolding transitions of the Trp-cage and Protein G fast-folding proteins. Multiple competing pathways have been identified that are consistent with earlier reports. The method generates physically meaningful transition pathways on a picosecond time scale (10-100 ps) for these systems. Furthermore, we demonstrate that these pathways serve as excellent initial seeds for path sampling methods, e.g., weighted ensemble (WE) simulations, substantially accelerating the convergence of rate constants compared to standard WE protocols. The PathGennie software is freely available.