GLM/PLM Embeddings

BRCA VEP (Evo2)

BRCA sequence variants are represented by Evo2 embeddings of their reference and altered DNA sequences, averaged over each sequence and joined together. The target assigns each variant to its recorded functional-effect class: FUNC, LOF, or INT.

ClassificationLicence: Source terms apply; redistribution authorized by dataset curatorlocal-BRCA_Evo2-VEPTarget: ylocal ↗

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)	0.4715	0.3606	0.4628	0.7723
Logistic Regression	0.4514	0.2978	0.4385	0.6982
RealMLP	0.4423	0.2668	0.4351	0.6601
RealTabPFN 2.5	0.437	0.3588	0.4334	0.7741
TabFM	0.4167	0.2654	0.4139	0.7371
Causilo	0.405	0.2727	0.4071	0.7459
MLP	0.4027	0.2017	0.4086	0.6378
Random Forest	0.3986	0.2151	0.3961	0.6482
TabM	0.3965	0.1524	0.398	0.5801
TabDPT	0.3944	0.1814	0.3903	0.7245
AutoGluon	0.3936	0.1841	0.3899	0.7228
LimiX2	0.3919	0.1641	0.407	0.6507
Extra Trees	0.386	0.2041	0.3876	0.6448
TabPFN 3.5	0.371	0.2367	0.3866	0.7523
Logistic Regression (tuned)	0.3473	0.1022	0.3779	0.5834
XGBoost	0.3437	0.1071	0.3609	0.6323
MLP (tuned)	0.3279	0.07647	0.3598	0.5386
RealTabPFN v2	0.3268	0.1583	0.357	0.7032
XGBoost (tuned)	0.3213	0.06096	0.3519	0.5432
Extra Trees (tuned)	0.3172	0.04981	0.3503	0.5379
Random Forest (tuned)	0.3034	0.0449	0.3445	0.5309
TabPFN Wide 5k (ne3)	0.2922	0.06136	0.3394	0.73
MITRA	0.2879	0.001217	0.3337	0.5538
GeneICL	0.1364	0.02651	0.3484	0.5163