New directions for diffusion-based network prediction of protein function: incorporating pathways with confidence

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein-protein interaction (PPI) networks are crucial for understanding cellular functions.
• Incorporating network noise, edge directions, and pathway information can enhance PPI network utility.
• Previous methods lacked a straightforward approach to integrate these complex network properties.

Purpose of the Study:

• To develop and evaluate novel methods for augmenting protein-protein interaction (PPI) network representations.
• To improve functional inference accuracy by integrating network noise, edge directions, and pathway information.
• To introduce extensions of diffusion state distance (DSD) that naturally incorporate these properties.

Main Methods:

• Defined three incremental versions of diffusion state distance (DSD): cDSD, caDSD, and capDSD.
• The capDSD matrix integrates confidence scores, directed edges, and pathway information.
• Tested four function prediction methods using these matrices on yeast PPI networks and STRING database associations.

Main Results:

• The capDSD matrix demonstrated superior performance in functional inference tasks.
• Weighted majority vote combined with capDSD yielded the best results on the Baker's yeast PPI network.
• Augmented DSD methods, particularly capDSD, outperformed models that independently incorporated pathway edges.

Conclusions:

• Augmented diffusion state distance (DSD) metrics provide a powerful framework for analyzing protein-protein interaction (PPI) networks.
• Integrating confidence, directionality, and pathways probabilistically enhances functional inference.
• The developed capDSD method offers a significant advancement in leveraging complex PPI network data.