Conformational Sampling for Transition State Searches on a Computational Budget

Transition state searches are the basis for computa-tionally characterizing reaction mechanisms, making them a pivotaltool in myriad chemical applications. Nevertheless, common searchalgorithms are sensitive to reaction conformations, and theconformational spaces of even medium-sized reacting systems aretoo complex to explore with brute force. Here, we show that it ispossible to train a classifier to learn the features of reactionconformers that conduce successful transition state searches, suchthat optimal conformers can be down-selected before incurring thecost of a high-level transition state search. The efficacy andtransferability of this approach were tested using four distinctbenchmarks comprising over three hundred individual reactions.Neglecting conformer contributions led to qualitatively incorrectactivation energy estimations for a broad range of reactions, whereas simple random forest classifiers reliably down-selected low-barrier reaction conformers for unseen reactions. The robust performance of these machine learning classifiers mitigates cost as afactor when implementing conformational sampling into contemporary reaction prediction workflows and opens up many avenues for further improvements as transition state data grow.

Automated summary

In this study, a classifier was trained to learn the features of reaction conformers that lead to successful transition state searches, enabling the down-selection of optimal conformers before incurring the cost of a high-level transition state search. The results showed that neglecting conformer contributions can lead to qualitatively incorrect activation energy estimations, while machine learning classifiers reliably down-select low-barrier reaction conformers.