CLTAP: A TAP-binding peptide prediction method using pre-trained large language-embedding models with contrastive learning to enhance contextual co-attention mechanism

The transporter associated with antigen processing (TAP) transports peptides to the endoplasmic reticulum, which is a crucial step for determining CD8⁺ T cell epitopes. CLTAP is a novel method for predicting TAP-binding peptides, which enhances prediction accuracy by integrating contrastive learning (CL) with a contextual co-attention mechanism. To the best of our knowledge, this is the first study to utilize CL for predicting TAP-binding peptides. Leveraging CL to capture the consistency and inconsistency among similar and dissimilar TAP-binding peptides, respectively, significantly enhanced the model’s ability to differentiate between various samples. Additionally, the contextual co-attention mechanism introduced in this study effectively integrates and emphasizes two distinct features of peptide sequences. CL further enhanced the fused features, resulting in discriminative representations and significantly improving the accuracy of TAP-binding peptide prediction. By integrating these two technologies, CLTAP demonstrates superior predictive performance than existing methods across multiple benchmark datasets. Evaluations on test datasets indicate that CLTAP outperforms the second-best method by 6.32 % in Accuracy, 15.26 % in Matthews Correlation Coefficient, 4.48 % in F1-score, and 4.13 % in AUROC. It offers an efficient and reliable tool for TAP-binding peptide research and applications. CLTAP is freely available at https://github.com/ChenWeiCCZU/CLTAP.