Metagenomics

Gut microbiome markers (healthy vs cirrhosis)

Stool metagenomes from the Qin liver cirrhosis study, represented as the presence or absence of microbial genetic markers in MetAML. The benchmark predicts whether a sample came from a participant with cirrhosis or a healthy control.

Binary classificationLicence: CC-BY-4.0gut-cirrhosisTarget: diseasemetagenomics ↗

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
Logistic Regression (tuned)	0.8783	0.7636	0.8795	0.9457
Kumo Tabular (M)	0.8745	0.7534	0.8753	0.9445
MLP	0.8734	0.7532	0.8744	0.9355
MLP (tuned)	0.8648	0.738	0.8665	0.9442
TabPFN 3.5	0.8613	0.7262	0.8617	0.9337
TabDPT	0.861	0.728	0.8621	0.9421
Logistic Regression	0.8607	0.7358	0.8632	0.9468
GeneICL	0.8598	0.7352	0.863	0.9393
RealMLP	0.8575	0.7196	0.8587	0.9442
MITRA	0.8575	0.7196	0.8585	0.9312
Extra Trees	0.8571	0.7194	0.8581	0.9353
Random Forest (tuned)	0.8566	0.7237	0.8587	0.9283
TabM	0.8564	0.7182	0.8572	0.9228
TabFM	0.8562	0.7215	0.857	0.9477
TabPFN Wide 5k (ne3)	0.853	0.7073	0.8532	0.9278
Extra Trees (tuned)	0.8526	0.7128	0.8539	0.9254
Random Forest	0.8484	0.7016	0.8494	0.9307
RealTabPFN v2	0.8482	0.7006	0.8485	0.9241
RealTabPFN 2.5	0.8442	0.6907	0.8445	0.9344
AutoGluon	0.8413	0.6977	0.8439	0.9424
Causilo	0.8397	0.6835	0.84	0.9195
LimiX2	0.8389	0.686	0.8398	0.9196
XGBoost	0.7997	0.6081	0.8008	0.8724
XGBoost (tuned)	0.7898	0.5925	0.7916	0.8607