Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CoTox: Chain-of-Thought-Based Molecular Toxicity Reasoning and Prediction

Jueon Park, Yein Park, Minju Song, Soyon Park, Donghyeon Lee, Seungheun Baek, Jaewoo Kang

TL;DR

CoTox introduces a chain-of-thought-based toxicity reasoning framework that fuses chemical structure information encoded as IUPAC names with biology-derived context (pathways and GO terms) to predict six organ-specific toxicities. Using prompting strategies and GPT-4o (and other LLMs), CoTox demonstrates improved predictive performance over traditional ML/DL baselines on the UniTox benchmark, while providing interpretable, mechanistic explanations. The approach highlights the importance of linguistic-friendly chemical representations and context-aware reasoning for multi-organ toxicity assessment, including case studies that align with known toxicology literature. This work suggests a practical pathway for early-stage drug safety evaluation with interpretable AI that integrates structural and biological data in a unified reasoning framework.

Abstract

Drug toxicity remains a major challenge in pharmaceutical development. Recent machine learning models have improved in silico toxicity prediction, but their reliance on annotated data and lack of interpretability limit their applicability. This limits their ability to capture organ-specific toxicities driven by complex biological mechanisms. Large language models (LLMs) offer a promising alternative through step-by-step reasoning and integration of textual data, yet prior approaches lack biological context and transparent rationale. To address this issue, we propose CoTox, a novel framework that integrates LLM with chain-of-thought (CoT) reasoning for multi-toxicity prediction. CoTox combines chemical structure data, biological pathways, and gene ontology (GO) terms to generate interpretable toxicity predictions through step-by-step reasoning. Using GPT-4o, we show that CoTox outperforms both traditional machine learning and deep learning model. We further examine its performance across various LLMs to identify where CoTox is most effective. Additionally, we find that representing chemical structures with IUPAC names, which are easier for LLMs to understand than SMILES, enhances the model's reasoning ability and improves predictive performance. To demonstrate its practical utility in drug development, we simulate the treatment of relevant cell types with drug and incorporated the resulting biological context into the CoTox framework. This approach allow CoTox to generate toxicity predictions aligned with physiological responses, as shown in case study. This result highlights the potential of LLM-based frameworks to improve interpretability and support early-stage drug safety assessment. The code and prompt used in this work are available at https://github.com/dmis-lab/CoTox.

CoTox: Chain-of-Thought-Based Molecular Toxicity Reasoning and Prediction

TL;DR

CoTox introduces a chain-of-thought-based toxicity reasoning framework that fuses chemical structure information encoded as IUPAC names with biology-derived context (pathways and GO terms) to predict six organ-specific toxicities. Using prompting strategies and GPT-4o (and other LLMs), CoTox demonstrates improved predictive performance over traditional ML/DL baselines on the UniTox benchmark, while providing interpretable, mechanistic explanations. The approach highlights the importance of linguistic-friendly chemical representations and context-aware reasoning for multi-organ toxicity assessment, including case studies that align with known toxicology literature. This work suggests a practical pathway for early-stage drug safety evaluation with interpretable AI that integrates structural and biological data in a unified reasoning framework.

Abstract

Drug toxicity remains a major challenge in pharmaceutical development. Recent machine learning models have improved in silico toxicity prediction, but their reliance on annotated data and lack of interpretability limit their applicability. This limits their ability to capture organ-specific toxicities driven by complex biological mechanisms. Large language models (LLMs) offer a promising alternative through step-by-step reasoning and integration of textual data, yet prior approaches lack biological context and transparent rationale. To address this issue, we propose CoTox, a novel framework that integrates LLM with chain-of-thought (CoT) reasoning for multi-toxicity prediction. CoTox combines chemical structure data, biological pathways, and gene ontology (GO) terms to generate interpretable toxicity predictions through step-by-step reasoning. Using GPT-4o, we show that CoTox outperforms both traditional machine learning and deep learning model. We further examine its performance across various LLMs to identify where CoTox is most effective. Additionally, we find that representing chemical structures with IUPAC names, which are easier for LLMs to understand than SMILES, enhances the model's reasoning ability and improves predictive performance. To demonstrate its practical utility in drug development, we simulate the treatment of relevant cell types with drug and incorporated the resulting biological context into the CoTox framework. This approach allow CoTox to generate toxicity predictions aligned with physiological responses, as shown in case study. This result highlights the potential of LLM-based frameworks to improve interpretability and support early-stage drug safety assessment. The code and prompt used in this work are available at https://github.com/dmis-lab/CoTox.

Paper Structure

This paper contains 16 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Toxicity Prompt Construction
  5. LLM Inference with Toxicity Prompting
  6. Experiments
  7. Benchmark Dataset
  8. Experimental settings
  9. Results & Discussion
  10. Prompt-wise Toxicity Prediction Performance
  11. Model-wise Toxicity Prediction Performance
  12. Impact of Structural Representations: SMILES vs. IUPAC
  13. Case Study: Reasoning Outcome Analysis
  14. Case Study: Organ-Specific Cell Line Guided Reasoning
  15. Conclusion
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Overview of CoTox Framework, (A) Toxicity Prompt Construction, where toxicity-related pathways and GO terms are retrieved from the CTD and additional extracting toxicity-related information is carried out by GPT-4o, followed by integration with IUPAC for toxicity prediction. (B) LLM inference process using a structured toxicity prompt to generate reasoning and predictions based on extracted biological data and chemical structure.
  • Figure 2: Comparison of Structural reasoning using SMILES vs IUPAC input.