GLM/PLM Embeddings

DeepLoc2 Fungi (ESM-2)

Fungal proteins from the DeepLoc 2.0 training and validation collection are represented by ESM-2 sequence embeddings, averaged across amino-acid residues. The task classifies each protein as membrane-associated or soluble using the source annotations.

Binary classificationLicence: DeepLoc 2.0 source terms; ESM-2 model MIT; redistribution authorized by dataset curatorlocal-DeepLoc2-Fungi-ESM2Target: membranelocal ↗

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
Kumo Tabular (M)	0.7693	0.546	0.7598	0.8338
GeneICL	0.7637	0.5527	0.736	0.8376
TabFM	0.7612	0.5541	0.729	0.8339
MLP (tuned)	0.7609	0.5286	0.7474	0.8117
Causilo	0.7601	0.5484	0.7334	0.8273
Logistic Regression (tuned)	0.7588	0.5336	0.7631	0.8413
LimiX2	0.7568	0.518	0.7656	0.8294
TabPFN 3.5	0.7521	0.5281	0.7255	0.8311
Logistic Regression	0.7519	0.5212	0.7319	0.8203
MITRA	0.7513	0.519	0.7321	0.82
TabDPT	0.7511	0.5095	0.7502	0.8163
RealTabPFN 2.5	0.7477	0.5126	0.7288	0.8266
TabPFN Wide 5k (ne3)	0.7416	0.5245	0.7116	0.8333
RealTabPFN v2	0.7409	0.4988	0.7221	0.8164
AutoGluon	0.7409	0.5069	0.7293	0.8175
RealMLP	0.7197	0.487	0.6993	0.8081
MLP	0.6967	0.4446	0.6965	0.8005
TabM	0.6944	0.4392	0.6795	0.8145
Random Forest (tuned)	0.6655	0.4133	0.6392	0.7857
Extra Trees (tuned)	0.6597	0.4072	0.6342	0.7961
Random Forest	0.5914	0.3169	0.5842	0.7841
XGBoost	0.5531	0.2234	0.567	0.7291
XGBoost (tuned)	0.5461	0.2077	0.562	0.7018
Extra Trees	0.5318	0.252	0.5493	0.7904