Improvements in chemical reaction pathway exploration algorithms and dataset generation

Abstract

Chemical reaction networks provide a comprehensive framework for understanding complex reaction systems, in which reaction path exploration is a critical component. In this study, molecular structures are represented as bond-electron matrices, and reaction candidates are systematically enumerated through matrix transformations. Starting from more than 1,000 reactant molecules, diverse reaction pathways were generated and validated using DFT calculations, resulting in OrgReact, a dataset comprising 9,649 reactions. The dataset includes reactant, product, and transition-state structures, together with associated energetic information, and is intended to support data-driven studies of organic reaction pathways and machine learning models for molecular energies and forces.