Retro-Expert: Collaborative Reasoning for Interpretable Retrosynthesis

Retrosynthesis prediction aims to infer the reactant molecules based on a given product molecule, which is a fundamental task in chemical synthesis. However, existing methods rely on a static pattern-matching paradigm, which limits their ability to perform effective logical decision-making from chemical data, leading to a black-box process. We propose Retro-Expert, an interpretable retrosynthesis framework that performs collaborative reasoning by combining the complementary strengths of Large Language Models and specialized models via pure reinforcement learning. It outputs natural language explanations grounded in chemical logic through three components: (1) specialized models provide chemical knowledge that is distilled into a high-quality chemical decision space, (2) LLM-driven critical reasoning to generate predictions with an interpretable reasoning path, and (3) knowledge-grounded policy optimization refines the interpretable decision policy. Experiments show that Retro-Expert surpasses both LLM-based and specialized models across different metrics, while generating chemically grounded explanations that enhance chemists' trust in practice. The source code for this paper is available at https://github.com/MagixRab-ll/Retro-Expert

Retrosynthesis is the task of finding the reactants needed to make desired product molecule, which is important for drug discovery and chemical synthesis. However, existing methods work like black boxes: they often predict reactants by memorizing reaction patterns from training data, rather than reasoning through chemical principles. We introduce Retro-Expert, a framework that makes retrosynthesis prediction more accurate and transparent. Retro-Expert combines two kinds of models: (1) Specialized models first provide useful chemical clues, such as possible reaction types, centers, and reactant candidates. (2) A large language model then reasons over these clues, selects plausible candidates, or rejects and corrects incorrect ones, and explains its decision in natural language. We further train the model with pure reinforcement learning so that it learns to make more reliable chemistry-grounded decisions. Retro-Expert improves prediction accuracy while also producing faithful explanations that chemists can inspect, which make AI-assisted synthesis planning more useful for chemical discovery.