Other

PRISM Stool Metabolomics CD vs UC

Untargeted stool metabolomics measurements from the PRISM inflammatory bowel disease cohort, using the study's C18 negative-ion mass spectrometry data. The task is to distinguish Crohn's disease from ulcerative colitis using metabolite feature abundances.

Binary classificationLicence: CC BY 4.0MW-ST001000-CD-UCTarget: diagnosismetabolomics ↗

At this operating point

Model comparison

p=10,000, n=50 · 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
Extra Trees	0.7198	0.4785	0.7295	0.7965
Extra Trees (tuned)	0.6904	0.42	0.7052	0.7811
Kumo Tabular (M)	0.6827	0.4002	0.6968	0.7903
GeneICL	0.6785	0.3769	0.6871	0.7648
TabPFN 3.5	0.6746	0.3817	0.6881	0.7746
RealMLP	0.6741	0.4013	0.6933	0.7538
Random Forest	0.6671	0.3734	0.6861	0.7841
Logistic Regression	0.6661	0.3451	0.676	0.7347
TabM	0.6608	0.3544	0.6713	0.7515
Causilo	0.6606	0.3818	0.6781	0.7069
RealTabPFN 2.5	0.6603	0.3586	0.6766	0.7545
MLP (tuned)	0.6557	0.3293	0.6646	0.773
MLP	0.6525	0.317	0.6598	0.7715
TabPFN Wide 5k (ne3)	0.6519	0.3418	0.6672	0.7527
Random Forest (tuned)	0.6508	0.3308	0.6638	0.7716
MITRA	0.6508	0.3353	0.6653	0.7181
TabFM	0.6494	0.3146	0.6546	0.7374
RealTabPFN v2	0.6254	0.2778	0.6377	0.7161
TabDPT	0.6214	0.327	0.6486	0.7121
LimiX2	0.6204	0.2728	0.6346	0.7265
XGBoost (tuned)	0.5754	0.2322	0.6046	0.7114
XGBoost	0.545	0.1746	0.5824	0.6603
Logistic Regression (tuned)	0.4875	0.1184	0.563	0.7529