Chemistrees: data driven identification of reaction pathways via machine learning

TL;DR

This paper introduces a supervised machine learning approach using decision trees to identify key geometric features in molecular dynamics data that predict state transitions, providing an unbiased analysis of reaction pathways.

Contribution

The study presents a novel application of decision trees to analyze molecular dynamics, enabling unbiased identification of reaction mechanisms without relying on chemical intuition.

Findings

Successfully identified proton transfer mechanisms in water clusters

Demonstrated effectiveness of decision trees in analyzing molecular trajectories

Provided mechanistic insights into proton exchange reactions

Abstract

We propose a supervised machine learning algorithm, decision trees, to analyze molecular dynamics output. The approach aims to identify the predominant geometric features which correlate with trajectories that transition between two arbitrarily defined states. The data-based algorithm aims to identify such features in an approach which is unbiased by human "chemical intuition". We demonstrate the method by analyzing proton exchange reactions in formic acid (FA) solvated in small water clusters. The simulations were performed with ab initio molecular dynamics combined with a method for generating rare events, specifically path sampling. Our machine learning analysis identified mechanistic descriptions of the proton transfer reaction for the different water clusters.