M17: NOVEL QUANTITATIVE METHOD FOR GENETIC ASSOCIATION TESTING

Abstract

Background Traditional genome-wide association studies perform a per-SNP association test, resulting in millions of tests. To subsequently examine effects at higher levels, gene-based, pathway-based, or polygenic risk approaches are used to aggregate the SNP-level association results. These methods result in a high multiple-testing burden, are vulnerable to inflation due to effects of Linkage Disequilibrium (LD) and gene size, and require time-expensive computing per phenotype. We propose a quantitative scoring method that operates directly on SNP-level data and can be used for any arbitrary genetic region of interest. We hypothesize that (a) our method can robustly identify genetic regions of interest, that (b) our method can be used to explain variance in a similar manner to polygenic risk approaches, and that (c) our method is robust against effects of LD and gene size. These properties ensure that our novel method can be used for genetic association testing. Methods Our primary data set consists of the Nijmegen Biomedical Study: 4452 genome-wide genotyped subjects using the Illumina HumanOmniExpress-12 and -24 BeadChip platforms with available Body Mass Index (BMI) measurements. In this data set, we tested the effects of LD by incremental pruning, the effects of gene size by Kolmogorov-Smirnov distribution comparison, and we compared our association results to results obtained using existing genetic association applications (Plink 1.9, Magma 1.04, and PRSice 1.25). Secondly, we compare variance explained for different methods in an Attention Deficit Hyperactivity Disorder (ADHD) discovery (n=2947) and replication (n=785) cohort. Results Using our novel method and fewer than 4500 individuals, we find one significantly associated gene for BMI (SNRPC) and several other suggestive genes which are confirmed in literature to be associated with BMI and were not picked up by the other methods tested. Secondly, we find a similar variance explained for ADHD across cohorts compared to the existing method. Lastly, our method is invariant to gene sizes and shows robust results against the effects of LD. Discussion Our results show that our novel method can identify true genetic regions of interest for BMI, can explain variance across cohorts for ADHD, and proves robust against the effects of gene size and LD. We identified genes of interest that were missed by existing methods, suggesting that our method could add to existing genetic association tests. More work in larger cohorts is needed to identify in which precise conditions our method can increase power to detect genetic regions of interest, however these first results in both a physical and a psychiatric/behavioral phenotype show promise.