Benchmarking of protein interaction databases for integration with manually reconstructed signalling network models.

Area of Science:

• Bioinformatics
• Systems Biology
• Molecular Biology

Background:

• Protein interaction databases are vital for network bioinformatics and integrating molecular data.
• Their utility in constructing predictive computational models of biological networks requires evaluation.

Purpose of the Study:

• To benchmark five protein interaction databases (X2K, Reactome, Pathway Commons, Omnipath, Signor) for recovering manually curated interactions.
• To assess their performance in reconstructing logic-based network models of cardiac hypertrophy, mechano-signalling, and fibrosis.

Main Methods:

• Benchmarking protein interaction databases against manually curated edges from three biological network models.
• Evaluating the databases' ability to recover interactions, including tissue-specific and transcriptional regulation.
• Testing the capacity of Signor and Pathway Commons to identify novel interactions that enhance model predictions.

Main Results:

• Pathway Commons demonstrated the best performance in recovering interactions across all three network models (71% for hypertrophy, 68% for mechano-signalling, 69% for fibroblast).
• Databases successfully recovered well-conserved pathways but performed poorly on tissue-specific and transcriptional regulation, indicating a knowledge gap.
• Signor and Pathway Commons identified new interactions, including the role of Ca2+/calmodulin-dependent protein kinase II phosphorylation of CREB in cardiomyocyte hypertrophy.

Conclusions:

• Protein interaction databases are valuable for network reconstruction but have limitations in capturing tissue-specific and regulatory details.
• Manual curation remains critical for building comprehensive and accurate biological network models.
• The study provides a framework for database benchmarking and reveals novel insights into cardiac hypertrophy signalling pathways.