DeepYoke/ToxiMol-benchmark

ToxiMol: A Benchmark for Structure-Level Molecular Detoxification

Overview

ToxiMol is the first comprehensive benchmark for molecular toxicity repair tailored to general-purpose Multimodal Large Language Models (MLLMs). This is the dataset repository for the paper "Breaking Bad Molecules: Are MLLMs Ready for Structure-Level Molecular Detoxification?".

Key Features

🧬 Comprehensive Dataset

• 560 representative toxic molecules spanning diverse toxicity mechanisms and varying granularities
• 11 primary toxicity repair tasks based on Therapeutics Data Commons (TDC) platform
• Multi-granular coverage: Tox21 (12 sub-tasks), ToxCast (10 sub-tasks), and 9 additional datasets
• Multimodal inputs: SMILES strings + 2D molecular structure images rendered using RDKit

🎯 Challenging Task Definition

The molecular toxicity repair task requires models to:

1. Identify potential toxicity endpoints from molecular structures
2. Interpret semantic constraints from natural language descriptions
3. Generate structurally similar substitute molecules that eliminate toxic fragments
4. Satisfy drug-likeness and synthetic feasibility requirements

📊 Systematic Evaluation

• ToxiEval framework: Automated evaluation integrating toxicity prediction, synthetic accessibility, drug-likeness, and structural similarity
• Comprehensive analysis: Evaluation of ~30 mainstream MLLMs with ablation studies
• Multi-dimensional metrics: Success rate analysis across different evaluation criteria and failure modes

Dataset Structure

Task Overview

Dataset	Task Type	# Molecules	Description
AMES	Binary Classification	50	Mutagenicity testing
Carcinogens	Binary Classification	50	Carcinogenicity prediction
ClinTox	Binary Classification	50	Clinical toxicity data
DILI	Binary Classification	50	Drug-induced liver injury
hERG	Binary Classification	50	hERG channel inhibition
hERG_Central	Binary Classification	50	Large-scale hERG database with integrated cardiac safety profiles
hERG_Karim	Binary Classification	50	hERG data from Karim et al.
LD50_Zhu	Regression (log(LD50) < 2)	50	Acute toxicity
Skin_Reaction	Binary Classification	50	Adverse skin reactions
Tox21	Binary Classification (12 sub-tasks)	60	Nuclear receptors, stress response pathways, and cellular toxicity mechanisms (ARE, p53, ER, AR, etc.)
ToxCast	Binary Classification (10 sub-tasks)	50	Diverse toxicity pathways including mitochondrial dysfunction, immunosuppression, and neurotoxicity

Data Structure

Each entry contains:

{
  "task": "string",      // Toxicity task identifier
  "id": "int",           // Molecule ID
  "smiles": "string",    // SMILES representation
  "image": "binary" // 2D molecular structure image binary
}

Available Subdatasets

subdatasets = [
    "ames", "carcinogens_lagunin", "clintox", "dili", 
    "herg", "herg_central", "herg_karim", "ld50_zhu", 
    "skin_reaction", "tox21", "toxcast"
]

# Load all datasets
datasets = {}
for name in subdatasets:
    datasets[name] = load_dataset("DeepYoke/ToxiMol-benchmark", data_dir=name)

Experimental Results

Our systematic evaluation of ~30 mainstream MLLMs reveals:

• Current limitations: Overall success rates remain relatively low across models
• Emerging capabilities: Models demonstrate initial potential in toxicity understanding, semantic constraint adherence, and structure-aware molecule editing
• Key challenges: Structural validity, multi-dimensional constraint satisfaction, and failure mode attribution

Citation

If you use this dataset in your research, please cite:

@misc{lin2025breakingbadmoleculesmllms,
      title={Breaking Bad Molecules: Are MLLMs Ready for Structure-Level Molecular Detoxification?}, 
      author={Fei Lin and Ziyang Gong and Cong Wang and Yonglin Tian and Tengchao Zhang and Xue Yang and Gen Luo and Fei-Yue Wang},
      year={2025},
      eprint={2506.10912},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2506.10912}, 
}

License

This project is licensed under the MIT License - see the LICENSE file for details.