Molecular properties

Acute Toxicity LD50 (Zhu)

The Zhu dataset contains acute oral toxicity measurements in rats. The benchmark predicts pLD50 from molecular fingerprints: the negative base-10 logarithm of the median lethal dose, measured in moles per kilogram of body weight.

RegressionLicence: CC BY 4.0 via TDC; underlying source terms require verification before re-hostingTDC-LD50-ZhuTarget: Ytdc ↗

At this operating point

Model comparison

p=10,000, n=100 · Lower RMSE 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	RMSE	MAE	R²
TabPFN 3.5	0.8673	0.6549	0.1781
GeneICL	0.8874	0.6771	0.1376
XGBoost (tuned)	0.8961	0.6913	0.1184
LimiX2	0.9038	0.6817	0.1078
Extra Trees	0.9038	0.6815	0.1079
Random Forest	0.9045	0.6822	0.1065
RealTabPFN 2.5	0.906	0.6909	0.1012
Logistic Regression	0.9062	0.695	0.1005
Kumo Tabular (M)	0.9104	0.6972	0.09195
MITRA	0.9114	0.6883	0.09276
MLP	0.9156	0.7061	0.07865
AutoGluon	0.9167	0.6968	0.07961
TabFM	0.9182	0.6981	0.07449
Random Forest (tuned)	0.9184	0.6947	0.07224
TabDPT	0.9196	0.7103	0.0729
TabM	0.9213	0.7051	0.07248
Logistic Regression (tuned)	0.9229	0.7038	0.06457
Extra Trees (tuned)	0.9279	0.7064	0.05412
RealTabPFN v2	0.9283	0.7079	0.05406
MLP (tuned)	0.93	0.723	0.0452
XGBoost	0.9461	0.7255	0.02012
RealMLP	0.9707	0.736	-0.03297
Causilo	1.004	0.7668	-0.1024