Determining effects of non-synonymous SNPs on protein-protein interactions using supervised and semi-supervised...

Area of Science:

• Genetics and Bioinformatics
• Computational Biology
• Molecular Interactions

Background:

• Single nucleotide polymorphisms (SNPs) are common genetic variations linked to complex disorders.
• Non-synonymous missense SNPs (nsSNPs) near protein-protein interaction (PPI) interfaces can alter protein binding.
• Predicting the impact of nsSNPs on PPIs is crucial but challenging.

Purpose of the Study:

• To develop a computational method, SNP-IN, for predicting the effects of nsSNPs on PPIs.
• To classify nsSNP effects as strengthening/weakening, disrupting/preserving, or detrimental/neutral/beneficial.
• To assess the utility of SNP-IN in analyzing disease-centered networks.

Main Methods:

• Developed SNP-IN, a predictor utilizing supervised and semi-supervised classifiers.
• Trained classifiers on mutagenesis data from 151 PPI complexes with experimental binding affinities.
• Evaluated 11 classifiers on three classification problems, including a Random Forest self-learning protocol.

Main Results:

• SNP-IN achieved promising performance, with weighted f-measures from 0.87 (2-class) to 0.70 (3-class).
• Integrating 2-class predictions improved the 3-class classifier's accuracy.
• Demonstrated SNP-IN's utility in studying nsSNP-induced rewiring of disease networks.

Conclusions:

• SNP-IN provides an accurate and balanced method for predicting nsSNP effects on PPIs.
• The tool facilitates the functional annotation of disease-associated SNPs.
• SNP-IN is accessible as a web server for large-scale network analysis.