Unsupervised\nMachine Learning in the Analysis of Nonadiabatic\nMolecular Dynamics Simulation

The all-atomic full-dimensional-level\nsimulations of\nnonadiabatic\nmolecular dynamics (NAMD) in large realistic systems has received\nhigh research interest in recent years. However, such NAMD simulations\nnormally generate an enormous amount of time-dependent high-dimensional\ndata, leading to a significant challenge in result analyses. Based\non unsupervised machine learning (ML) methods, considerable efforts\nwere devoted to developing novel and easy-to-use analysis tools for\nthe identification of photoinduced reaction channels and the comprehensive\nunderstanding of complicated molecular motions in NAMD simulations.\nHere, we tried to survey recent advances in this field, particularly\nto focus on how to use unsupervised ML methods to analyze the trajectory-based\nNAMD simulation results. Our purpose is to offer a comprehensive discussion\non several essential components of this analysis protocol, including\nthe selection of ML methods, the construction of molecular descriptors,\nthe establishment of analytical frameworks, their advantages and limitations,\nand persistent challenges.