Accelerating All-Atom Simulations and Gaining Mechanistic Understanding of Biophysical Systems through State Predictive Information Bottleneck

An effective implementation of enhanced sampling algorithms for molecular dynamics simulations requiresa prioriknowledge of the approximate reaction coordinate describing therelevant mechanisms in the system. In this work, we focus on therecently developed artificial intelligence-based State PredictiveInformation Bottleneck (SPIB) approach and demonstrate howSPIB can learn such a reaction coordinate as a deep neural networkeven from undersampled trajectories. We exemplify its usefulnessby achieving more than 40 times acceleration in simulating twomodel biophysical systems through well-tempered metadynamicsperformed by biasing along the SPIB-learned reaction coordinate. These include left- to right-handed chirality transitions in asynthetic helical peptide (Aib)9and permeation of a small benzoic acid molecule through a synthetic, symmetric phospholipidbilayer. In addition to significantly accelerating the dynamics and achieving back and forth movement between different metastablestates, the SPIB-based reaction coordinate gives mechanistic insights into the processes driving these two important problems