2025/12/21 by Pinku Nath, Y. Ono, Yu Harabuchi +4 · 1 voice
Materials Science · Computer Science · Chemistry · #Machine Learning in Materials Science #Computational Drug Discovery Methods #Asymmetric Hydrogenation and Catalysis
paper · doi:10.1021/acscatal.5c06298
openalex created_date 2025/12/21 · openalex publication_date 2025/12/21 · openalex updated_date 2026/02/08
ChemOntology is a computational framework developed to extract and apply chemical knowledge from reaction intermediates generated during automated reaction path searches. By integrating chemical and geometric knowledge generated using chemical ontology and topology, ChemOntology identifies chemically relevant reaction paths and geometries to guide the reaction path search. Combined with an automated reaction path search method, Artificial Force Induced Reaction (AFIR) and applied to the Heck reaction, the ChemOntology-AFIR approach efficiently identified all key intermediates and elementary steps, including major and side products, even in high-energy regions where conventional AFIR need much larger computational efforts for extensive conformational sampling. This knowledge-driven approach significantly reduces computational costs by eliminating chemically irrelevant paths and structures. ChemOntology relies on three main inputs: the reaction setup, chemically informed assumptions encoded as Elementary Reaction Process Ontologies (ERPOs), and a set of reaction rules. Together, these enable efficient use of extracted knowledge to accelerate reaction exploration. Unlike machine learning models, it requires no training on data sets and is broadly applicable to a wide range of organometallic systems, offering a robust tool for mechanistic analysis and rational catalyst design, especially in systems where human insight remains indispensable.