Molecular properties

Tox21 Aryl Hydrocarbon Receptor Activity

This Tox21 task records compound activity in the aryl hydrocarbon receptor signalling pathway. The benchmark uses molecular fingerprints to predict whether a compound receives an active or inactive label in the AhR assay.

Binary classificationLicence: Tox21 / MoleculeNet distribution; underlying source terms applyMoleculeNet-Tox21-NR-AhRTarget: NR-AhRmoleculenet ↗

At this operating point

Model comparison

p=10,000, n=100 · Higher Macro-F1 is better

Random Forest is marked with a diamond. Models are ordered by the selected metric.

Across sample budgets

Learning curves

Models

All metric values at this operating point

Model	Macro-F1	MCC	Balanced accuracy	ROC-AUC
LimiX2	0.5925	0.1928	0.5912	0.7361
TabFM	0.5814	0.1998	0.5668	0.6968
Causilo	0.5682	0.1612	0.5628	0.7059
Kumo Tabular (M)	0.5609	0.1695	0.5517	0.7334
RealMLP	0.5528	0.1661	0.5455	0.6937
RealTabPFN 2.5	0.5504	0.1551	0.5446	0.7263
TabDPT	0.5478	0.1341	0.5424	0.627
AutoGluon	0.5378	0.1057	0.5473	0.6813
MLP (tuned)	0.5264	0.1087	0.5383	0.6004
GeneICL	0.5188	0.1197	0.5248	0.6898
RealTabPFN v2	0.5095	0.09029	0.5201	0.7243
TabPFN 3.5	0.5028	0.09489	0.5163	0.7269
Extra Trees (tuned)	0.5022	0.08033	0.5249	0.6987
Logistic Regression	0.4991	0.09659	0.5145	0.6932
XGBoost	0.4734	0.02268	0.5024	0.6583
XGBoost (tuned)	0.4734	0.02268	0.5024	0.6158
Random Forest	0.4726	0.04583	0.502	0.7044
Extra Trees	0.4725	0.03694	0.502	0.7081
TabM	0.4718	0.008471	0.5012	0.6785
MITRA	0.4717	0.007507	0.5011	0.7321
MLP	0.4712	0.03054	0.5013	0.6128
Random Forest (tuned)	0.4699	0.0153	0.5007	0.7143
TabPFN Wide 5k (ne3)	0.4699	0.0153	0.5007	0.7049
Logistic Regression (tuned)	0.4697	0.01167	0.5004	0.7076