Pep2Net: an integrated peptide knowledge base and cross-attention model for multi-task prediction of peptides

Abstract

Pep2Net is a multi-label peptide property prediction framework that combines pre-trained ESM-2 protein language model embeddings with handcrafted physicochemical features through cross-attention and low-rank bilinear pooling. This record archives version 1.0.0 of the source code used for the manuscript “Pep2Net: an integrated peptide knowledge base and cross-attention model for multi-task prediction of peptides”. The public development repository is available at https://github.com/xiaoyao00a/Pep2Net.