Prioritization of Cancer-Related Genomic Variants by SNP Association Network

We have developed a general framework to construct an association network of single nucleotide polymorphisms (SNPs) (SNP association network, SAN) based on the functional interactions of genes located in the flanking regions of SNPs. SAN, which was constructed based on protein–protein interactions in the Human Protein Reference Database (HPRD), showed significantly enriched signals in both linkage disequilibrium (LD) and long-range chromatin interaction (Hi-C). We used this network to further develop two methods for predicting and prioritizing disease-associated genes from genome-wide association studies (GWASs). We found that random walk with restart (RWR) using SAN (RWR-SAN) can greatly improve the prediction of lung-cancer-associated genes by comparing RWR with the use of network in HPRD (AUC 0.81 vs 0.66). In a reanalysis of the GWAS dataset of age-related macular degeneration (AMD), SAN could identify more potential AMD-associated genes that were previously ranked lower in the GWAS study. The interactions in SAN could facilitate the study of complex diseases.