AI Framework Synthegy Enhances Chemical Synthesis Planning
- Synthegy, developed at EPFL, uses large language models (LLMs) to rank synthesis routes against chemist-defined goals, aligning with expert judgments in 71.2% of cases.
- The framework was validated through a double-blind study involving 36 independent chemists across 368 evaluations.
- Senior researchers showed higher agreement with Synthegy’s selections compared to PhD students, indicating the system’s alignment with experienced chemists’ strategic intuitions.
- Evaluating 60 candidate routes with Synthegy takes approximately 12 minutes and costs about $2–3 in API fees.
- The code and benchmarks for Synthegy are publicly available on GitHub.
Synthegy offers a modular AI framework that enhances chemical synthesis planning by using LLMs to evaluate synthesis routes rather than generating molecules directly. This approach allows for broad applications in drug discovery and material design, as it aligns closely with expert chemists’ decisions.
The system’s ability to match expert judgments over seventy percent of the time demonstrates its potential in streamlining complex chemical processes and reducing laboratory errors. (Source)
