Information assessment on predicting protein-protein interactions.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Identifying protein-protein interactions is crucial for understanding cellular mechanisms.
• High-throughput experimental methods for detecting these interactions have limitations, including high false positive and negative rates.
• Integrating diverse genomic data can improve prediction accuracy and biological insights.

Purpose of the Study:

• To assess the contribution of different genomic features to predicting protein-protein interactions.
• To compare the effectiveness of various computational models for interaction prediction.
• To identify the most informative data types for robust protein-protein interaction prediction.

Main Methods:

• Utilized a Bayesian network approach to model protein-protein interactions genome-wide in yeast.
• Analyzed the predictive power of genomic features including mRNA expression, localization, essentiality, and functional annotation.
• Compared Bayesian networks with alternative models like logistic regression and random forests.

Main Results:

• Functional classification data (MIPS, Gene Ontology) significantly contributed to predicting protein-protein interactions.
• In a complete information subset, functional similarity was more informative than expression correlations or essentiality.
• Random forests and logistic regression showed potential to be more effective than Bayesian networks.

Conclusions:

• MIPS and Gene Ontology functional similarity datasets are dominant contributors for predicting protein-protein interactions.
• Random forests using only MIPS and GO data achieved high classification accuracy.
• Adding other genomic data provided minimal improvement in the complete information subset, and Bayesian discretization reduced performance.