An improved approach to infer protein-protein interaction based on a hierarchical vector space model

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene Ontology (GO) annotations are crucial for comparing gene product functions in biomedical research.
• Semantic similarity derived from GO is a key indicator for protein interactions.
• Existing vector space models for semantic similarity have limitations.

Purpose of the Study:

• To introduce a novel Hierarchical Vector Space Model (HVSM) for computing semantic similarity between genes.
• To enhance the basic vector space model by incorporating GO term relationships (is_a, part_of).
• To improve the accuracy of protein-protein interaction prediction.

Main Methods:

• Developed HVSM, which considers GO term ancestors and descendants.
• Introduced a Certainty Factor to calibrate semantic similarity based on gene annotations.
• Applied HVSM to Homo sapiens and Saccharomyces cerevisiae protein-protein interaction datasets.

Main Results:

• HVSM demonstrated significant improvement in distinguishing positive from negative protein interactions compared to existing methods.
• Achieved AUC score improvements of up to 12% over basic VSM and 11% over SimGIC.
• HVSM showed comparable or superior performance to other similarity measures in CESSM and TCSS tests.

Conclusions:

• HVSM offers a robust and improved approach for calculating semantic similarity in gene product function analysis.
• The method enhances the understanding of protein-protein interactions through advanced GO term utilization.
• HVSM software and supplementary information are publicly available.