← All datasets

Molecular properties

Blood-Brain Barrier Penetration (Martins)

The Martins dataset labels compounds by whether they cross the blood-brain barrier. The benchmark uses molecular fingerprints to distinguish compounds labelled as penetrating the barrier from those labelled as non-penetrating.

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

Mean cross-validation scores for this dataset. Each score comes from the requested feature- and sample-cap cell, and failed fits use the constant predictor.

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

All metric values at this operating point
ModelMacro-F1MCCBalanced accuracyROC-AUC
RealMLP0.72750.47510.70480.7641
RealTabPFN 2.50.71590.50980.68220.7669
TabDPT †0.71280.4610.69040.7679
TabICL v20.70740.43120.69640.7877
TabFM0.70290.43970.68420.7816
AutoGluon0.69990.46860.67480.7877
MITRA0.6960.48810.66510.7706
Logistic Regression0.68890.46880.6610.7793
MLP0.68830.4330.67090.7379
TabPFN Wide (8k) ‡0.68620.45850.65820.7781
TabPFN 30.68250.47030.65420.789
RealTabPFN v20.67880.45960.6510.7747
TabM0.6760.45480.64920.7885
TabPFN Wide 5k (ne3) ‡0.67420.45870.64690.7833
Random Forest0.6690.45630.64250.7789
Extra Trees0.6630.44770.63820.7831
CatBoost0.65980.44080.63570.7559
XGBoost0.65760.40310.63950.735
LightGBM0.63150.32210.61660.7023
KNN0.57960.33450.5790.692