MODEL_CARD.md

Model card - tox21_chemprop_classifier

Model details

• Model name: Chemprop Message Passing Neural Network Tox21 Baseline
• Developer: MIT (trained by JKU Linz)
• Paper URL: https://pubs.acs.org/doi/full/10.1021/acs.jcim.3c01250
• Model type / architecture:
  • Message Passing Neural Network implemented using Chemprop.
  • Hyperparameters: link to config – A multitask network is trained for all Tox21 targets.
• Inference: Access via FastAPI endpoint. Upon receiving a Tox21 prediction request, the model generates and returns predictions for all Tox21 targets simultaneously.
• Model version: v0
• Model date: 14.10.2025
• Reproducibility: Code for full training is available and enables retraining of the model from scratch.

Intended use

This model serves as a baseline benchmark for evaluating and comparing toxicity prediction methods across the 12 pathway assays of the Tox21 dataset. It is not intended for clinical decision-making without experimental validation.

Metric

Each Tox21 task is evaluated using the area under the receiver operating characteristic curve (AUC). Overall performance is reported as the mean AUC across all individual tasks.

Training data

Tox21 training and validation sets.

Evaluation data

Tox21 test set.