An Agentic Retrobiosynthesis Framework with Learned Frontier Selection
2026-08-31 • Computation and Language
Computation and LanguageArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors studied how well large language models (LLMs) can choose which molecules to explore first when tracing biochemical reaction paths backward, a process called retrobiosynthesis. They used a fixed biochemical system that generates possible reactions but varied the model's strategy for picking the next molecule to expand. They found that fine-tuned LLM policies solved more pathways with fewer attempts compared to other search methods, showing that improving the search strategy helps even when the underlying chemistry rules stay the same. However, the success still depends on how the possible next steps are built and ranked.
large language modelsretrobiosynthesisrule-based systemsEscherichia colisearch policyfine-tuningfrontier selectionreaction rankingMCTSbiochemical pathways
Authors
Philippe Meyer, Guillaume Gricourt, Thomas Duigou, Joan Hérisson, Jean-Loup Faulon
Abstract
Large language models are increasingly used as agents for multistep retrosynthesis, raising the question of how much their search policy contributes independently of the underlying reaction model. We investigate this question in a biological setting through rule-based retrobiosynthesis: a deterministic biochemical engine generates the same validated transitions for every method, searching for routes that terminate in metabolites available to an \emph{Escherichia coli} chassis, while the policy only selects which frontier molecule to expand next. Prompted and LoRA-tuned Qwen2.5-7B policies use a strict choice-only interface. The fine-tuned policy reaches $65\pm1$\% solve rate at 10 expansions on LASER versus 59\% for MCTS, and at 200 expansions reaches $78\pm1$\% versus 75\% on LASER, $88\pm3$\% versus 80\% on the RetroPath RL Golden benchmark, and $63\pm2$\% versus 45\% on the BioNavi-NP benchmark. Fine-tuning also consistently outperforms direct prompting. These results show that route-supervised frontier selection can improve budgeted search without altering biochemical generation, although performance remains dependent on frontier construction and reaction ranking.