A Multi-Modal Framework for Phage-Host Interaction Prediction Using Multi-View Contrastive Learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Antibiotic resistance is a critical global health challenge.
• Phage therapy offers a promising alternative to antibiotics.
• Accurate prediction of phage-host interactions (PHI) is essential for effective phage therapy.

Purpose of the Study:

• To develop a novel computational framework, MCL-PHI, for high-precision prediction of PHI.
• To address limitations in current methods regarding data integration and network representation.

Main Methods:

• Integration of multi-modal features and multi-view contrastive learning.
• Utilizing Large Language Models (LLMs) with Chain-of-Thought for knowledge description generation.
• Employing BioBERT for domain-enhanced semantic representation and genomic feature fusion.
• Constructing a heterogeneous network and applying a multi-view graph neural network with contrastive learning.
• Using an adaptive multi-head attention mechanism for link prediction.

Main Results:

• MCL-PHI significantly outperforms six state-of-the-art baseline methods in PHI prediction.
• Ablation studies confirm the effectiveness of individual framework components.
• Case analyses demonstrate the framework's potential for discovering novel PHIs.

Conclusions:

• MCL-PHI represents a significant advancement in computational prediction of phage-host interactions.
• The framework's ability to integrate diverse data and learn complex relationships enhances phage therapy development.
• MCL-PHI holds promise for accelerating the discovery of new phage-host pairs to combat antibiotic resistance.