Prediction of transition state structures of gas-phase chemical reactions via machine learning

Accurately predicting transition state (TS) structures is crucial for understanding chemical reaction mechanisms. This paper proposes a machine learning (ML) model that predicts TS structures for general organic reactions by deriving interatomic distances from atomic pair features of reactant, product, and linearly interpolated structures. The model demonstrates high accuracy, especially for bond formation/breakage, with a 93.8% success rate for quantum chemical saddle point optimizations and 88.8% of optimizations having energy errors below 0.1 kcal/mol. The model's ability to explore multiple reaction paths is also demonstrated.

Sunghwan Choi