Generative and Contrastive Self-Supervised Learning for Virulence Factor Identification Based on Protein-Protein...

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Virulence factors (VFs) are key to pathogen invasion and host damage.
• Understanding VFs aids in developing anti-virulence strategies.
• Current VF identification methods often ignore protein-protein interaction (PPI) data and struggle with imbalanced datasets.

Purpose of the Study:

• To develop a novel framework for accurate virulence factor identification.
• To address the challenges of imbalanced data and the underutilization of PPI information in existing models.
• To improve the understanding of pathogenic mechanisms and identify new anti-virulence targets.

Main Methods:

• Proposed a Generative and Contrastive self-supervised learning framework for Virulence Factor identification (GC-VF).
• Transformed VF identification into an imbalanced node classification task on graphs derived from PPI networks.
• Implemented a generative attribute reconstruction module for feature learning and a local contrastive learning module for capturing local features and context.

Main Results:

• GC-VF demonstrated superior performance compared to baseline methods on naturally imbalanced datasets.
• The framework achieved higher accuracy and stability in virulence factor identification.
• The study highlights the importance of incorporating PPI information and handling data imbalance.

Conclusions:

• GC-VF offers a robust solution for accurate virulence factor identification, especially in imbalanced data scenarios.
• The proposed framework enhances the understanding of pathogen mechanisms by leveraging PPI networks.
• This work provides a foundation for developing more effective anti-virulence therapies.