The 2nd SMLQC seminar (smlqc.mlatom.com) will be given by Max Pinheiro Jr on Feb. 16, 2023.
Discovering patterns in nonadiabatic molecular dynamics with machine learning: The ULaMDyn package
Owing to the high dimensionality ofNAMD-generated data, a typical challenging task is to find the key active coordinates that drive the molecular system through critical regions of the potential energy surfaces, thereby triggering chemical transformations. Also, the myriad of possible reaction pathways accessible in NAMD simulations adds an extra layer of difficulty to the data exploration problem. In this scenario, unsupervised machine learning (ML) can bring an automated solution for the in depth analysis of NAMD data, facilitating the interpretation and understanding of the underlying photo-dynamical processes. To contribute to this solution, we have developed the Unsupervised Learning Analysis of Molecular Dynamics (ULaMDyn) program that provides a complete data analysis pipeline, going from data curation to molecular representations, dimension reduction, and clustering analysis. The unsupervised learning methods implemented in ULaMDyn aim to surpass existing barriers for chemists to extract insights from NAMD simulations regardless of the complexity of the molecular system under study. In this talk, I will present the theoretical aspects of unsupervised learning followed by practical examples of dimensionality reduction and clustering techniques for analyzing NAMD data. We expect that the development of a new program based on unsupervised ML will pave the way for conceptual breakthroughs in the understanding of photochemical phenomena, as these methodologies provide an objectively improved analysis tool for discovering patterns in excited-states molecular dynamics without requiring prior knowledge of the underlying chemical reaction mechanisms.