Skip to main content

Table 16 Performance comparison of the models with different self-attention setup

From: Protein–protein interaction site prediction by model ensembling with hybrid feature and self-attention

Model

ACC

Pre

Rec

F1

MCC

AUROC

AUPRC

Base-model (4-head attention)

0.549

0.489

0.565

0.512

0.105

0.571

0.554

SH-model (Single-head attention)

0.528

0.454

0.540

0.491

0.056

0.539

0.528

NA-model (No attention)

0.522

0.317

0.549

0.375

0.051

0.540

0.543