Explainable Deep Hypergraph Learning Modeling the Peptide Secondary Structure Prediction.

Area of Science:

• Structural Biology
• Computational Biology
• Bioinformatics

Background:

• Predicting peptide secondary structures is crucial but challenging, especially for short peptides lacking sufficient discriminative information.
• Understanding secondary structures is vital for peptide tertiary structure reconstruction and functional analysis.

Purpose of the Study:

• To introduce PHAT, a novel deep hypergraph learning framework for accurate peptide secondary structure prediction.
• To develop an interpretable model that enhances understanding of structural feature representations and substructure classification.
• To demonstrate the utility of secondary structure prediction in downstream tasks like functional analysis.

Main Methods:

• Developed PHAT, a deep hypergraph learning framework incorporating a multi-head attention network.
• Integrated sequential semantic information from biological corpora and structural semantic information from multi-scale segmentation.
• Employed residue-based reasoning for enhanced structure prediction accuracy and interpretability.

Main Results:

• Achieved high accuracy in peptide secondary structure prediction, particularly for extremely short peptides.
• The interpretable models successfully highlighted reasoning for structural feature representations and secondary substructure classification.
• Demonstrated the importance of predicted secondary structures in peptide tertiary structure reconstruction and functional analysis.

Conclusions:

• PHAT offers a versatile and accurate solution for peptide secondary structure prediction, even with limited sequence data.
• The framework's interpretability provides insights into structural feature reasoning, aiding biological understanding.
• PHAT is expected to facilitate functional peptide design and contribute to advancements in structural biology research.