Risk quantification of multigenic conditions for SNP array based direct-to-consumer genomic services

Abstract

Genome wide association studies (GWAS) are typically designed as case-control studies, collecting thousands of sick and healthy individuals, genotyping hundreds of thousands of SNPs, and documenting the SNPs which are more abundant in one group or the other. Direct-to-consumer genetic testing has opened the possibility for a regular person to receive data about his/her genotype, but the validity of risk assessment procedures and the final genetic risk estimate have been questioned. Many authors have discussed the advantage of use of the asymptotic Bayes factor (ABF) to measure the strength of SNP/trait associations, over the use of p-values. We propose a ABF based heuristic to filterour and select SNP/trait associations to be used in multigenic risk assessment. A raw genotype result from the 23andMe web service was merged with the GWAS catalog, and SNP/trait associations were filtered and selected using the R programming language together with free and publicly available databases. From the initial 3195 SNP/trait associations, only 425 remained after the initial filters on descent, replicated findings, qualitative trait and availability of the number of cases and controls in the study. Selecting only one SNP/trait association from repeated studies and studies done with proxy SNPs left us with 377 SNP/trait associations available for multigenic risk assessment. After excluding the associations with unsatisfying ABF, only 300 SNP/trait associations remain for the multigenic risk assessment. Whatever the link between SNP/trait associations and final DTC multigenic risk assessment for a given trait is, the final value of a risk score is heavily influenced by the number, as well as strength of evidence for individual SNP/trait pairs that are used for calculation. The ABF provides an unambiguous and simple criterion for ranking and including SNP/trait associations in multigenic risk assessment.