gTPS: A machine learning and quantum computer-based algorithm for Transition Path Sampling

Simulating rare structural rearrangements of macromolecules with classical computational methods, such as Molecular Dynamics (MD), is an outstanding challenge. A multitude of technological advancements, from development of petaFLOPS supercomputers to advent of various enhance sampling methods, has granted access to time intervals of microseconds and even milliseconds in recent years. Yet, many key events occur on exponentially longer timescales. Here, path sampling techniques have the advantage of focusing the computational power on barrier-crossing trajectories, but generating uncorrelated transition paths that explore significantly different conformational regions remains a problem. To address this issue, we devised a hybrid path-sampling scheme, graph-Transition Path Sampling (gTPS), that generates the trial transition pathways using a quantum annealer. We first employ a classical computer to perform an uncharted exploration of the conformational space using a data-driven MD method. The dataset is then post-processed using a path-integral-based method to obtain a coarse-grained network representation of reactive pathways. By resorting to quantum annealing, the entire ensemble of these pathways can be encoded into a superposition in the initial quantum state of the annealer. Finally, by performing the quantum adiabatic transition on the state of the annealer, one can potentially generate/sample uncorrelated paths while they retain a high statistical probability (follow low free energy regions). We have first validated this scheme on a prototypically simple transition (α_R↔C_5 of alanine dipeptide) which could be extensively characterized on a desktop computer. Subsequently, we scaled up in complexity by generating a protein conformational transition (Bovine Pancreatic Trypsin Inhibitor - BPTI) that occurs on the millisecond timescale, obtaining results that match those of the Anton special-purpose supercomputer. Finally, we dicuss our current investigations on the application of gTPS to the unfolding process of headpiece subdomain of Villin and BPTI. Despite limitations due to the available quantum hardware, our study highlights how realistic biomolecular simulations provide a potentially impactful new ground for applying, testing, and advancing quantum technologies.