Synthegy AI Guides Molecule Design

Researchers at École Polytechnique Fédérale de Lausanne, led by Philippe Schwaller and Andres M Bran, have developed Synthegy, a new AI framework that uses large language models (LLMs) to guide traditional computational chemistry tools. Synthegy addresses two major challenges in chemistry: retrosynthesis and reaction mechanisms. Instead of directly generating chemical structures, Synthegy evaluates and steers existing software by interpreting chemical strategies expressed in natural language. This allows chemists to articulate their goals in plain language, leading to more efficient and strategically aligned molecular design. In double-blind expert studies, Synthegy's assessments aligned with human judgments over 70% of the time. The technology is expected to accelerate drug discovery and improve reaction design, making advanced computational tools more accessible. Collaborators include National Centre of Competence in Research Catalysis and B12 Labs, with funding from Intel and Merck Group. The study was published in Matter (journal).