Predicting Protein Functions Based on Differential Co-expression and Neighborhood Analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Functional Genomics

Background:

• Proteins are crucial for biological processes, with interactions extending beyond physical binding.
• Existing methods for predicting protein function often overlook diverse intrinsic information within feature and label spaces.
• Integrating protein interaction and gene expression data is key for identifying context-specific biological networks.

Purpose of the Study:

• To develop a novel integrative method, CNPFP, for accurate protein function prediction.
• To identify yeast cell cycle-specific proteins by linking differentially expressed proteins within interaction networks.
• To exploit intrinsic and latent linkages in heterogeneous data using genomic features.

Main Methods:

• Developed CNPFP, integrating differential co-expression analysis and a neighbor-voting algorithm.
• Employed a global iterative approach to exploit heterogeneous data and genomic features.
• Utilized a global iterative neighbor-voting algorithm for relevant feature subset selection.

Main Results:

• Identified eight condition-specific modules, including a subnetwork enriched with cyclin-dependent kinases crucial for cell cycle regulation.
• Provided comprehensive annotations for 3538 Saccharomyces cerevisiae proteins.
• Achieved high performance metrics: AUROC of 0.9862, accuracy of 0.9710, and F-score of 0.9691.

Conclusions:

• Integrating diverse data and exploiting intrinsic protein relationships improves protein function prediction quality.
• The CNPFP method demonstrates significant utility in functional genomics studies.
• The findings highlight the importance of considering inherent data characteristics for robust biological network analysis.