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

Using molecular dynamics (MD) simulations to observe rare events and generating the mechanistic pathways for such events, e.g. ligand (un)binding or protein (un)folding, is computationally challenging due to the prolonged waiting times associated with such transitions over high free-energy barriers or through entropic bottlenecks. Traditional enhanced sampling techniques typically rely on bias potentials or elevated temperatures to facilitate such transitions, and the unbiased path sampling methods may be computationally expensive. In this work, we introduce a new idea towards rapid generation of transition pathways for rare events using a computationally efficient direction-guided adaptive sampling protocol that preferentially selects trajectories progressing toward the desired goal, which may be either to escape from a local minimum (e.g. drug unbinding) or reaching a specific target (e.g. protein folding with known target structure) in a pre-defined progress variable space. This approach leverages the idea that rare events are typically associated with short transition time, but very long waiting time. Our method attempts to quickly identify such transition pathways using ultrashort (tens of femtoseconds), rapidly monitored trajectories (samplers) initiated with randomized velocities, selectively propagating those exhibiting desired movement in a collective variable (CV) space. Demonstrating broad applicability and effectiveness, we apply this framework to diverse rare event scenarios, including multiple ligand unbinding pathways in T4 Lysozyme and Abl kinase proteins, and (un)folding events in fast-folding proteins (Trp-cage and Protein G). Our results show that this approach dramatically accelerates pathway generation (within a few hundred picoseconds) while staying close to the minimum free energy pathways providing viable starting points for other advanced path sampling methods like transition path sampling (TPS) and weighted ensemble (WE) simulations. Moreover, this method is able to identify multiple competing pathways with an approximate estimate of their relative population. The code associated with this protocol, PathGennie, is freely available at https://github.com/TeamSuman/PathGennie.