MFP-MFL: Leveraging Graph Attention and Multi-Feature Integration for Superior Multifunctional Bioactive Peptide...

Area of Science:

• Biochemistry and Bioinformatics
• Computational Biology
• Genomics

Background:

• Bioactive peptides are crucial for biological functions but are difficult to classify due to their multifunctionality.
• Accurate prediction of peptide function is essential for their effective utilization in research and development.

Purpose of the Study:

• To present MFP-MFL, an advanced multi-feature, multi-label learning framework for predicting multifunctional peptides.
• To enhance the accuracy and robustness of peptide function prediction using integrated deep learning models.

Main Methods:

• Integration of Graph Attention Networks (GAT) with protein language models (ESM-2, ProtT5, RoBERTa).
• Application of an ensemble learning strategy to leverage deep sequence features and functional dependencies.
• Validation through comparative experiments and a large-scale mutation case study.

Main Results:

• MFP-MFL achieved high performance metrics: precision (0.799), coverage (0.821), and accuracy (0.786).
• The model demonstrated strong predictive capability with an Absolute True score of 0.737 and a low Absolute False score of 0.086.
• A case study on 86,970 mutations confirmed the model's ability to predict functional changes due to sequence variations.

Conclusions:

• MFP-MFL is a powerful and accurate tool for the discovery and application of multifunctional peptides.
• The framework offers significant potential to advance research in peptide science and biomedical applications.
• The integration of GAT and protein language models provides a robust approach for complex biological predictions.