Abstract
BACKGROUND: Therapeutic peptides have become an important direction in drug discovery because of their high targeting and low side effects, and are used to treat many diseases. Peptides are short-chain molecules formed by connecting amino acids through peptide bonds and play key roles in the body. The stability and production costs of peptides are challenges that need to be overcome for their pharmaceutical applications. Researchers have improved the accuracy of therapeutic peptide sequence function predictions by constructing and integrating peptide features from different sources. However, accurately predicting multi-functional therapeutic peptides is challenging due to the limitations of handcrafted feature properties, which are unable to capture the full complexity of biological systems. RESULTS: In this study, we introduce a novel method TPpred-CMvL for the prediction of multi-functional therapeutic peptide (MTP) based on a contrastive multi-view learning model. This framework directly integrates semantic information pretraining TAPE from protein large language model and evolutionary information. Subsequently, TPpred-CMvL leverages contrastive multi-view learning to comprehensively capture representations of peptide sequences, thereby enhancing the prediction accuracy of MTPs. We utilized adaptive synthetic sampling and focal loss to address the classification imbalance arising from the long-tailed distribution. The experimental results demonstrate that the proposed method outperforms existing related approaches and exhibits the most advanced performance. CONCLUSION: We developed a contrast multi-view learning model TPpred-CMvL utilizing sequential semantic information TAPE and evolutionary information PSSM. Compared with existing related methods, this method achieved state-of-the-art performance. Finally, a web server has been established and is accessible at http://bliulab.net/TPpred-CMvL .