An automated analytical framework for identifying reaction channels and structural evolution in multi-channel nonadiabatic dynamics: Combining dimensionality reduction, clustering, and information entropy

The automatic identification of reaction channels and key nuclear motions from nonadiabatic dynamics simulations remains a major challenge. Traditional manual analysis is inefficient and relies heavily on expert intuition, creating a bottleneck for interpreting complex photochemical processes. To overcome this challenge, we introduce an automated analytical framework that integrates unsupervised learning (dimensionality reduction, clustering, and information entropy) with conical intersection validation to directly extract key dynamical information from on-the-fly trajectory surface hopping data. Our method effectively determines multi-channel reaction pathways and successfully identifies their key structural nuclear motions. When applied to keto-isocytosine and methaniminium cation, the framework successfully recovers all known reaction channels and their characteristic coordinates, demonstrating its reliability. This work provides an effective, objective, and reproducible approach for transforming raw trajectory data into clear mechanistic insights in excited-state dynamics.