Mapping Gene Ontology to proteins based on protein-protein interaction data

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene Ontology (GO) provides a standardized vocabulary for gene and protein function annotation across organisms.
• Protein-protein interaction (PPI) datasets are crucial for understanding protein roles and cellular mechanisms.
• Integrating GO and PPI data enables more accurate prediction of functions for uncharacterized proteins.

Purpose of the Study:

• To develop and evaluate a computational method for predicting yeast protein function.
• To leverage protein-protein interaction data and Gene Ontology annotations for functional prediction.
• To assess the prediction accuracy using a rigorous validation approach.

Main Methods:

• Application of a Markov random field model to predict protein function.
• Utilizing multiple protein-protein interaction datasets from MIPS (physical and genetic interactions).
• Employing leave-one-out cross-validation and a functional path matching scheme for evaluation.

Main Results:

• The method assigns a confidence probability to predicted protein functions across GO categories (cellular component, molecular function, biological process).
• Achieved 52% precision and recall for biological process prediction using leave-one-out validation.
• Outperformed simple guilty-by-association methods in predicting yeast protein function.

Conclusions:

• The Markov random field approach effectively integrates GO and PPI data for robust protein function prediction.
• The developed method offers a significant improvement over existing techniques for yeast functional genomics.
• This approach enhances the annotation of uncharacterized proteins, advancing our understanding of cellular processes.