AMHF-TP: Multifunctional therapeutic peptides prediction based on multi-granularity hierarchical features.

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Multifunctional therapeutic peptides (MFTPs) show great therapeutic promise but are difficult to predict using traditional methods.
• Existing methods face challenges like long training times, small datasets, and poor generalization.

Purpose of the Study:

• To develop an advanced computational method for accurate and robust recognition of MFTPs.
• To overcome the limitations of traditional MFTP identification techniques.

Main Methods:

• Proposed AMHF-TP, incorporating migration learning, CNNs, self-attention, hypergraph construction, and hierarchical feature extraction.
• Leveraged pretrained models for atomic compositional features and refined extraction from amino acid sequences and secondary structures.
• Integrated multimodal peptide sequence features for comprehensive analysis.

Main Results:

• AMHF-TP demonstrated superior precision, accuracy, and coverage compared to leading methods.
• Comparative analysis confirmed the exceptional performance and stability of AMHF-TP in MFTP recognition tasks.
• The combined model outperformed separate hierarchical models and five contemporary methods.

Conclusions:

• AMHF-TP offers an effective and robust solution for MFTP recognition.
• The method's advanced feature extraction and integration capabilities enhance predictive performance.
• This work advances the identification of potential therapeutic peptides for drug discovery.