Chemists Can Now Design Molecules by Describing Them in Plain Language
A new AI system called Synthegy lets chemists guide molecular design using everyday language instead of navigating complex computational interfaces. Researchers at EPFL developed the framework by combining traditional chemistry algorithms with large language models that evaluate and rank proposed solutions.
The system addresses two persistent bottlenecks in chemistry: retrosynthesis planning and reaction mechanism analysis. Both tasks normally require years of expertise to execute well.
How the System Works
Synthegy starts with a target molecule and a simple instruction written in natural language. A chemist might specify that a particular ring should form early in the synthesis or that unnecessary protective steps should be avoided.
Standard retrosynthesis software then generates many possible pathways to build that molecule. Synthegy converts each pathway into text and has a language model evaluate how well it matches the chemist's stated goals. The system scores and explains its reasoning, letting chemists quickly filter for the most promising routes.
The same approach applies to reaction mechanisms-the step-by-step movement of electrons that determines how reactions proceed. Synthegy breaks these down into basic components and steers the search toward pathways that make chemical sense.
Testing With Working Chemists
A double-blind study had 36 chemists evaluate 368 synthesis pathways. The system's assessments matched the chemists' judgments 71.2% of the time on average.
Larger language models performed better than smaller ones, suggesting that model size matters for chemical reasoning. The system could flag unnecessary protecting steps, judge reaction feasibility, and prioritize efficient solutions.
A Different Role for AI
Rather than replacing chemist decision-making, Synthegy positions language models as guides that interpret and refine computational results. Chemists describe their goals in plain language and receive solutions that reflect their strategy.
The approach could accelerate drug discovery and make advanced synthesis tools more accessible to researchers who lack deep expertise in retrosynthesis planning.
Learn more about how large language models power practical applications or explore AI for science and research.
Your membership also unlocks:
