ChemATP: A Training-Free Chemical Reasoning Framework for Large Language Models

2025-12-22

AI Summary

Large language models (LLMs) exhibit inaccurate reasoning in molecular science primarily due to standard string-based molecular representations lacking explicit, fine-grained chemical priors. Training-based approaches statically embed domain knowledge into model parameters—compromising generalizability and knowledge updatability—while training-free methods rely solely on coarse-grained prompts, failing to support atom-level reasoning. Method: We propose the first training-free chemical reasoning framework that decouples domain knowledge from the inference engine, enabling dynamic retrieval of atom-level chemical textual knowledge under a frozen LLM. Contribution/Results: Our approach introduces (1) the first atom-level structured chemical text knowledge base, and (2) a chemistry-aware RAG mechanism with interface adaptation for frozen LLMs. Evaluated across diverse chemical reasoning tasks, it achieves 12.7–34.1% absolute accuracy gains over training-free baselines, matches state-of-the-art trained models, and preserves interpretability, knowledge updatability, and general-purpose reasoning capability.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Large Language Models (LLMs) exhibit strong general reasoning but struggle in molecular science due to the lack of explicit chemical priors in standard string representations. Current solutions face a fundamental dilemma. Training-based methods inject priors into parameters, but this static coupling hinders rapid knowledge updates and often compromises the model's general reasoning capabilities. Conversely, existing training-free methods avoid these issues but rely on surface-level prompting, failing to provide the fine-grained atom-level priors essential for precise chemical reasoning. To address this issue, we introduce ChemATP, a framework that decouples chemical knowledge from the reasoning engine. By constructing the first atom-level textual knowledge base, ChemATP enables frozen LLMs to explicitly retrieve and reason over this information dynamically. This architecture ensures interpretability and adaptability while preserving the LLM's intrinsic general intelligence. Experiments show that ChemATP significantly outperforms training-free baselines and rivals state-of-the-art training-based models, demonstrating that explicit prior injection is a competitive alternative to implicit parameter updates.

Problem

• Enables LLMs to perform precise chemical reasoning without training
• Decouples chemical knowledge from reasoning engine for adaptability
• Provides atom-level priors to overcome limitations of surface-level prompting

Innovation

• Decouples chemical knowledge from reasoning engine
• Uses atom-level textual knowledge base for retrieval
• Enables dynamic reasoning with frozen LLMs