Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Mingxu Zhang, Dazhong Shen, Qi Zhang, Ying Sun

TL;DR

ChemATP tackles the lack of explicit chemical priors in LLM-driven chemistry tasks by decoupling knowledge from reasoning. It builds an atom-level textual knowledge base using AtomDisc tokens and RDKit-derived descriptors, then uses a three-stage pipeline—prior selection, top-k analogue retrieval, and integrative inference with explicit priors and exemplars—to ground LLM reasoning without updating model weights. Empirically it achieves state-of-the-art results among training-free baselines and rivals training-based methods on MoleculeNet benchmarks, while offering transparent, evidence-based reasoning. The framework demonstrates backbone-agnostic applicability and zero-shot robustness, suggesting explicit grounding can rival parametric updates for scientific reasoning.

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.

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

TL;DR

ChemATP tackles the lack of explicit chemical priors in LLM-driven chemistry tasks by decoupling knowledge from reasoning. It builds an atom-level textual knowledge base using AtomDisc tokens and RDKit-derived descriptors, then uses a three-stage pipeline—prior selection, top-k analogue retrieval, and integrative inference with explicit priors and exemplars—to ground LLM reasoning without updating model weights. Empirically it achieves state-of-the-art results among training-free baselines and rivals training-based methods on MoleculeNet benchmarks, while offering transparent, evidence-based reasoning. The framework demonstrates backbone-agnostic applicability and zero-shot robustness, suggesting explicit grounding can rival parametric updates for scientific reasoning.

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.
Paper Structure (64 sections, 2 equations, 2 figures, 9 tables)

This paper contains 64 sections, 2 equations, 2 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Molecule Representation Learning.
  4. Chemical Large Language Models.
  5. Retrieval Augmented Generation in Chemistry.
  6. Method
  7. Atomic Textualized Prior Knowledge Base
  8. Atom Tokenization.
  9. Knowledge Base Construction.
  10. Stage I: Prior Selection and Task Formulation
  11. Stage II: Retrieval of Top-k Analogues
  12. Stage III: Integrative Inference with Priors and Analogues
  13. Explicit priors as inductive bias.
  14. Analogues as anchors.
  15. Experiments
  16. ...and 49 more sections

Figures (2)

  • Figure 1: Overview of ChemATP. ChemATP is a training-free, three-stage framework in which an LLM (i) analyzes the prior schema and selects task-relevant atom- and molecule-level priors, (ii) retrieves top-$k$ analogue molecules via Morgan fingerprint similarity and attaches their priors and labels as exemplars, and (iii) integrates the query molecule’s priors, the retrieved analogues, and the feature description in a structured prompt to make the inference.
  • Figure 2: Comparison between SMILES-only CoT and ChemATP on a BACE-1 inhibition task. The left panel illustrates how the baseline falls into a hallucination trap, deriving an overconfident prediction from surface-level structural motifs without physical verification. In contrast, the right panel demonstrates ChemATP's evidence-based reasoning: by explicitly citing quantitative priors (e.g., TPSA, LogP), it successfully rejects the candidate based on physicochemical constraints. Red highlights denote the verifiable attributes guiding the correct decision.