Filtering high-throughput protein-protein interaction data using a combination of genomic features

Area of Science:

• Systems Biology
• Bioinformatics
• Computational Biology

Background:

• High-throughput experiments generate vast protein-protein interaction data.
• This data often contains numerous spurious or incorrect interactions.
• Validation is crucial for accurate molecular network construction and prediction.

Purpose of the Study:

• To develop a method for validating protein-protein interactions from high-throughput experiments.
• To assess the reliability of these interactions using genomic features.
• To estimate the proportion of true interactions in various model organisms.

Main Methods:

• Utilized a combination of three genomic features: Pfam domains, Gene Ontology annotations, and sequence homology.
• Employed Bayesian network approaches to assign reliability scores to protein-protein interactions.
• Applied the method to high-throughput data from Saccharomyces cerevisiae, Caenorhabditis elegans, Drosophila melanogaster, and Homo sapiens.

Main Results:

• Protein-protein interactions supported by genomic features showed higher likelihood ratios, indicating greater reliability.
• The developed method achieved 90% sensitivity and 63% specificity.
• 56% of high-throughput interactions in Saccharomyces cerevisiae were deemed highly reliable.
• Estimated true interaction proportions were 27% (C. elegans), 18% (D. melanogaster), and 68% (H. sapiens).

Conclusions:

• A combination of sequence, structure, and annotation data effectively predicts true protein-protein interactions in noisy datasets.
• The method provides a reliable way to assign likelihood ratios to interactions, enhancing data quality.
• This approach significantly improves the accuracy of molecular network analysis.