Inferring non-synonymous single-nucleotide polymorphisms-disease associations via integration of multiple similarity networks.

Detecting associations between human genetic variants and their phenotypic effects is a significant problem in understanding genetic bases of human-inherited diseases. The focus is on a typical type of genetic variants called non-synonymous single nucleotide polymorphisms (nsSNPs), whose occurrence may potentially alter the structures of proteins, affecting functions of proteins, and thereby causing diseases. Most of the existing methods predict associations between nsSNPs and diseases based on features derived from only protein sequence and/or structure information, and give no information about which specific disease an nsSNP is associated with. To cope with these problems, the identification of nsSNPs that are associated with a specific disease from a set of candidate nsSNPs as a binary classification problem has been formulated. A new approach has been adopted for predicting associations between nsSNPs and diseases based on multiple nsSNP similarity networks and disease phenotype similarity networks. With a series of comprehensive validation experiments, it has been demonstrated that the proposed method is effective in both recovering the nsSNP-disease associations and inferring suspect disease-associated nsSNPs for both diseases with known genetic bases and diseases of unknown genetic bases.