Molecular properties

CYP3A4 Inhibition (Veith)

This dataset records whether compounds inhibit CYP3A4, a human enzyme involved in drug metabolism. The benchmark uses molecular fingerprints to predict the inhibition labels from the screening study by Veith and colleagues.

Binary classificationLicence: CC BY 4.0 via TDC; underlying source terms require verification before re-hostingTDC-CYP3A4-VeithTarget: Ytdc ↗

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	0.6436	0.2925	0.6424	0.7104
TabM	0.63	0.2923	0.6323	0.7125
Logistic Regression	0.629	0.2877	0.6305	0.6966
Kumo Tabular (M)	0.629	0.2645	0.6281	0.6947
TabPFN 3.5	0.6287	0.2905	0.6305	0.7227
RealTabPFN 2.5	0.6256	0.2788	0.6268	0.7233
TabDPT	0.6251	0.2733	0.6264	0.6448
GeneICL	0.6241	0.2832	0.6269	0.7053
MLP	0.6214	0.2514	0.6207	0.658
Extra Trees (tuned)	0.6169	0.2547	0.6177	0.7056
MLP (tuned)	0.6164	0.2445	0.6171	0.6533
AutoGluon	0.6147	0.2429	0.6149	0.6748
Causilo	0.6094	0.2212	0.6114	0.6617
Extra Trees	0.6078	0.3024	0.6201	0.7493
TabFM	0.6056	0.2133	0.605	0.6509
Random Forest	0.6027	0.2934	0.6162	0.7463
Logistic Regression (tuned)	0.5901	0.2839	0.6119	0.7093
XGBoost	0.5799	0.2014	0.5907	0.6736
MITRA	0.5737	0.2429	0.6069	0.7402
RealMLP	0.5722	0.196	0.588	0.6472
XGBoost (tuned)	0.5697	0.2093	0.5889	0.672
RealTabPFN v2	0.5669	0.2274	0.5883	0.7119
Random Forest (tuned)	0.5536	0.2146	0.5883	0.7259
TabPFN Wide 5k (ne3)	0.4814	0.1808	0.5528	0.7413