AI Aids Chemists in Step-by-Step Molecule Design

Chemists are gaining a new tool to accelerate the design of complex molecules, as artificial intelligence systems begin to assist in planning step-by-step synthetic pathways with greater strategic insight. This development addresses long-standing challenges in chemistry where predicting viable reaction sequences and understanding molecular mechanisms have traditionally relied on years of expert intuition.

A study led by Philippe Schwaller at EPFL has introduced a framework called Synthegy that uses large language models not to generate chemical structures directly, but to evaluate and guide established computational tools in chemistry. By interpreting chemical strategies expressed in natural language, these AI systems help bridge the gap between algorithmic search capabilities and the nuanced reasoning of experienced chemists.

The approach tackles two persistent problems in modern chemistry: retrosynthesis, where chemists work backward from a target molecule to identify feasible building blocks and reaction pathways, and reaction mechanisms, which describe the step-by-step movement of electrons during chemical transformations. While computers can explore vast numbers of potential pathways, they often lack the strategic judgment to prioritize the most plausible or efficient routes.

According to Andres M Bran, the first author of the Synthegy paper published in Matter, the user interface is critical when designing tools for chemists. Previous computational aids relied on complex filters and rigid rules that were difficult to navigate, limiting their adoption in laboratory settings. Synthegy aims to overcome this by allowing researchers to interact with the system using natural language descriptions of chemical strategies.

This method builds on earlier work demonstrating that large language models can serve as effective intermediaries between users and specialized AI-driven molecular generators. An open-source chatbot named ChatChemTS, developed for this purpose, assists chemists in designing new molecules through simple conversational interactions, automating the construction of reward functions based on desired molecular properties.

In tests, ChatChemTS has supported de novo design cases involving chromophores and anticancer drugs, specifically epidermal growth factor receptor inhibitors, demonstrating its utility in both single-objective and multi-objective optimization scenarios. The tool is available as an open-source package on GitHub, enabling broader access for academic and industrial researchers.

Separate research highlighted in Nature confirms that AI-assisted programs are already contributing to tangible outcomes in the lab, with one system enabling the synthesis of 35 new compounds by helping researchers overcome a major bottleneck in chemical synthesis planning. These advances suggest that AI is transitioning from theoretical aid to practical instrument in molecular discovery.

As the volume of known chemical reactions continues to grow — with hundreds of thousands added annually — the ability to efficiently navigate this expanding knowledge space becomes increasingly vital for drug discovery and advanced materials development. By combining the scale of computational search with AI-driven interpretive reasoning, tools like Synthegy and ChatChemTS aim to make molecular design more accessible, efficient, and guided by chemical insight rather than brute-force enumeration.