Researchers at EPFL have developed Synthegy, an AI framework that lets chemists guide complex molecule synthesis using simple language instructions. The system combines traditional algorithms with large language models to evaluate and rank reaction pathways. It also aids in understanding reaction mechanisms, potentially speeding up drug discovery.
Creating complex molecules for drugs or materials traditionally demands years of expertise in retrosynthesis, where chemists work backward from a target compound to identify starting materials and reaction routes. Synthegy, developed by a team led by Philippe Schwaller at EPFL, changes this by allowing chemists to input natural language instructions, such as forming a ring early or avoiding protecting groups. Standard software generates pathways, which the AI then scores and explains for alignment with those goals, as described in a paper published in Matter. Andres M. Bran, the first author, said, 'With Synthegy, we're giving chemists the power to just talk, allowing them to iterate much faster and navigate more complex synthetic ideas.' The framework applies similar reasoning to reaction mechanisms, breaking them into electron movements and evaluating feasibility under specified conditions. In a double-blind study with 36 chemists providing 368 evaluations, the system's assessments matched human judgments 71.2% of the time. Larger language models excelled in analyzing functional groups and full routes. Bran added, 'The connection between synthesis planning and mechanisms is very exciting: we usually use mechanisms to discover new reactions that enable us to synthesize new molecules.' Contributors include the National Centre of Competence in Research Catalysis and b12 Labs. The journal reference is Andres M. Bran et al., Matter, 2026; DOI: 10.1016/j.matt.2026.102812.
