TPpred-LE: therapeutic peptide function prediction based on label embedding.

Area of Science:

• Biochemistry and Bioinformatics
• Drug Discovery and Development

Background:

• Therapeutic peptides are crucial in physiology, treatment, and biopharmacy.
• Existing computational methods for predicting peptide functions lack explicit exploitation of inter-functional relationships.
• The need for advanced computational methods is driven by the discovery of new peptide functions and limited labeled data.

Purpose of the Study:

• To propose a novel computational method for predicting multiple functions of therapeutic peptides.
• To explicitly extract function correlation information and exploit function-specific classifiers.
• To develop a strategy for detecting new therapeutic peptide functions with limited labeled data.

Main Methods:

• A novel method, TPpred-LE, based on the Transformer framework was developed.
• Label embedding methodology was used to extract function correlation information.
• Function-specific classifiers and multi-label classifier retraining (MCRT) were incorporated.

Main Results:

• TPpred-LE demonstrated superior performance compared to existing state-of-the-art methods.
• The integration of MCRT with TPpred-LE proved robust for predicting functions with limited labeled data.
• The study highlighted the significance of inter-functional relationships in therapeutic peptide function prediction.

Conclusions:

• TPpred-LE is an effective function-specific classifier for accurate therapeutic peptide function prediction.
• Exploiting relationship information among functions is vital for improving prediction accuracy.
• MCRT offers a simple yet effective approach for function detection with limited labeled data.