LARC: Towards Human-level Constrained Retrosynthesis Planning through an Agentic Framework

2025-08-15

AI Summary

Constrained retrosynthetic planning seeks feasible synthetic routes from commercially available starting materials under multiple practical constraints, yet conventional methods struggle to balance computational efficiency and chemical plausibility. This paper introduces LARC, a large language model–based framework featuring a tool-augmented agent system. LARC pioneers a “proxy-as-judge” mechanism that dynamically integrates constraint evaluation—encompassing reaction feasibility, reagent availability, and operational complexity—directly into the reasoning process, enabling real-time, multi-constraint-guided search. Evaluated on 48 retrosynthetic tasks spanning three constraint categories, LARC achieves a 72.9% success rate, substantially outperforming baseline models while requiring significantly less computational time than traditional approaches; its performance approaches that of human experts.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large language model (LLM) agent evaluators leverage specialized tools to ground the rational decision-making of LLMs, making them well-suited to aid in scientific discoveries, such as constrained retrosynthesis planning. Constrained retrosynthesis planning is an essential, yet challenging, process within chemistry for identifying synthetic routes from commercially available starting materials to desired target molecules, subject to practical constraints. Here, we present LARC, the first LLM-based Agentic framework for Retrosynthesis planning under Constraints. LARC incorporates agentic constraint evaluation, through an Agent-as-a-Judge, directly into the retrosynthesis planning process, using agentic feedback grounded in tool-based reasoning to guide and constrain route generation. We rigorously evaluate LARC on a carefully curated set of 48 constrained retrosynthesis planning tasks across 3 constraint types. LARC achieves a 72.9% success rate on these tasks, vastly outperforming LLM baselines and approaching human expert-level success in substantially less time. The LARC framework is extensible, and serves as a first step towards an effective agentic tool or a co-scientist to human experts for constrained retrosynthesis.

Problem

• Automating retrosynthesis planning with practical constraints
• Improving synthetic route identification using LLM agents
• Bridging performance gap between AI and human experts

Innovation

• LLM agent evaluators using specialized tools
• Agent-as-a-Judge for constraint evaluation
• Tool-based reasoning for route generation