Tox21 Leaderboard

Tox21 Leaderboard 🧪

Measuring AI progress in Drug Discovery

Rank	Type	Model	Organization	Publication	Avg. AUC	Avg. ΔAUC-PR	# Parameters	ROC-AUC												ΔAUC-PR
Rank	Type	Model	Organization	Publication	Avg. AUC	Avg. ΔAUC-PR	# Parameters	NR-AR	NR-AR-LBD	NR-AhR	NR-Aromatase	NR-ER	NR-ER-LBD	NR-PPAR-gamma	SR-ARE	SR-ATAD5	SR-HSE	SR-MMP	SR-p53	NR-AR	NR-AR-LBD	NR-AhR	NR-Aromatase	NR-ER	NR-ER-LBD	NR-PPAR-gamma	SR-ARE	SR-ATAD5	SR-HSE	SR-MMP	SR-p53
🥇	🔼	DeepTox	JKU Linz	DeepTox: Toxicity Prediction using Deep Learning	0.847	0.302		0.807	0.850	0.928	0.834	0.793	0.815	0.839	0.841	0.793	0.858	0.941	0.862	0.221	0.052	0.509	0.350	0.421	0.265	0.167	0.301	0.433	0.223	0.437	0.248
🥈	🔼	SNN	JKU Linz	Self-Normalizing Neural Networks	0.844	0.261	1.9M	0.852	0.918	0.897	0.789	0.809	0.814	0.838	0.784	0.813	0.828	0.937	0.849	0.236	0.098	0.446	0.189	0.317	0.225	0.171	0.242	0.219	0.323	0.448	0.223
🥉	⤵️	CheMeleon	MIT	Descriptor-based Foundation Models for Molecular Property Prediction	0.838	0.294	40M	0.829	0.873	0.913	0.808	0.804	0.784	0.821	0.789	0.827	0.830	0.948	0.826	0.205	0.055	0.522	0.272	0.420	0.286	0.136	0.261	0.234	0.458	0.511	0.174
4	🔼	RF	JKU Linz	Measuring AI Progress in Drug Discovery: A Reproducible Leaderboard for the Tox21 Challenge	0.829	0.299	40.1M	0.781	0.769	0.916	0.823	0.814	0.768	0.832	0.800	0.809	0.841	0.946	0.851	0.198	0.042	0.456	0.315	0.417	0.285	0.203	0.239	0.290	0.333	0.534	0.274
5	🔼	SNN Ensemble	Rasayan Labs Inc.		0.827	0.291	19M	0.803	0.874	0.916	0.772	0.806	0.744	0.828	0.792	0.817	0.818	0.923	0.827	0.382	0.067	0.538	0.218	0.382	0.248	0.169	0.244	0.291	0.287	0.487	0.176
6	🔼	XGBoost	JKU Linz	Measuring AI Progress in Drug Discovery: A Reproducible Leaderboard for the Tox21 Challenge	0.823	0.277	460.7K	0.735	0.804	0.912	0.822	0.813	0.789	0.771	0.810	0.818	0.824	0.945	0.827	0.131	0.072	0.479	0.295	0.404	0.228	0.139	0.268	0.314	0.250	0.536	0.206
7	⤵️	GROVER	Tencent AI Lab (finetuned by JKU Linz)	Self-Supervised Graph Transformer on Large-Scale Molecular Data	0.822	0.233	48.4M	0.847	0.881	0.914	0.818	0.767	0.734	0.815	0.794	0.772	0.779	0.919	0.827	0.166	0.087	0.460	0.166	0.350	0.124	0.131	0.254	0.145	0.380	0.391	0.143
8	🔼	Chemprop	MIT (trained by JKU Linz)	Measuring AI Progress in Drug Discovery: A Reproducible Leaderboard for the Tox21 Challenge	0.815	0.232	709K	0.839	0.861	0.893	0.767	0.818	0.767	0.772	0.748	0.788	0.805	0.914	0.813	0.302	0.065	0.433	0.108	0.333	0.137	0.067	0.267	0.210	0.284	0.445	0.131
9	🔼	GIN	MIT & Stanford (trained by JKU Linz)	Measuring AI Progress in Drug Discovery: A Reproducible Leaderboard for the Tox21 Challenge	0.810	0.244	154K	0.808	0.882	0.890	0.773	0.771	0.778	0.740	0.756	0.787	0.774	0.930	0.836	0.281	0.097	0.399	0.142	0.381	0.277	0.072	0.213	0.240	0.261	0.416	0.144
10	⤵️	TabPFN	PriorLabs (trained by JKU Linz)	Measuring AI Progress in Drug Discovery: A Reproducible Leaderboard for the Tox21 Challenge	0.807	0.262	86.9M	0.753	0.741	0.893	0.770	0.782	0.806	0.793	0.798	0.787	0.816	0.942	0.806	0.161	0.030	0.413	0.223	0.402	0.266	0.158	0.293	0.222	0.328	0.477	0.173
11	🔼	D-MPNN (chemprop)	Independent Researcher	D-MPNN Model for Toxicity Prediction on the Original Tox21 Benchmark	0.799	0.262	2.1M	0.797	0.818	0.884	0.809	0.767	0.768	0.761	0.797	0.736	0.733	0.914	0.797	0.296	0.056	0.428	0.217	0.329	0.151	0.199	0.315	0.202	0.250	0.521	0.184
12	0️⃣	GPT-OSS	OpenAI (inference by JKU Linz)	Measuring AI Progress in Drug Discovery: A Reproducible Leaderboard for the Tox21 Challenge	0.702	0.083	120B	0.558	0.702	0.824	0.708	0.688	0.692	0.676	0.707	0.646	0.763	0.732	0.724	0.016	0.018	0.233	0.077	0.077	0.090	0.049	0.121	0.060	0.076	0.114	0.060

Avg. AUC: Mean ROC-AUC across all 12 tasks

Avg. ΔAUC-PR: Mean ΔAUC-PR across all 12 tasks

Rank: based on Avg. AUC

Type: 0️⃣ Zero-shot | 1️⃣ Few-shot | ⤵️ Pre-trained | 🔼 Models trained from scratch