A custom correlation coefficient (CCC) approach for fast identification of multi-SNP association patterns in genome-wide SNPs data.

Complex diseases are often associated with sets of multiple interacting genetic factors and possibly with unique sets of the genetic factors in different groups of individuals (genetic heterogeneity). We introduce a novel concept of custom correlation coefficient (CCC) between single nucleotide polymorphisms (SNPs) that address genetic heterogeneity by measuring subset correlations autonomously. It is used to develop a 3-step process to identify candidate multi-SNP patterns: (1) pairwise (SNP-SNP) correlations are computed using CCC; (2) clusters of so-correlated SNPs identified; and (3) frequencies of these clusters in disease cases and controls compared to identify disease-associated multi-SNP patterns. This method identified 42 candidate multi-SNP associations with hypertensive heart disease (HHD), among which one cluster of 22 SNPs (six genes) included 13 in SLC8A1 (aka NCX1, an essential component of cardiac excitation-contraction coupling) and another of 32 SNPs had 29 from a different segment of SLC8A1. While allele frequencies show little difference between cases and controls, the cluster of 22 associated alleles were found in 20% of controls but no cases and the other in 3% of controls but 20% of cases. These suggest that both protective and risk effects on HHD could be exerted by combinations of variants in different regions of SLC8A1, modified by variants from other genes. The results demonstrate that this new correlation metric identifies disease-associated multi-SNP patterns overlooked by commonly used correlation measures. Furthermore, computation time using CCC is a small fraction of that required by other methods, thereby enabling the analyses of large GWAS datasets.