VIDraft/tox21-classifier

VIDRAFT Tox21 Classifier

A Hugging Face Space exposing a /predict API for the Tox21 Leaderboard.

• —Input: a list of SMILES strings.
• —Output: a nested dict {smiles: {target: probability}} over the 12 Tox21 endpoints.
• —Model: per-task XGBoost on rich molecular features — Morgan ECFP (2048) + MACCS (167)
• —RDKit 2D descriptors — trained on the official ml-jku/tox21 training split.

Developed by VIDRAFT (Aether PharmaOS). Reproducible: train.py retrains from the official dataset; inference featurization runs on-the-fly in predict.py.

API

POST /predict with {"smiles": ["CCO", "c1ccccc1"]} → {"predictions": {...}}.