Replication of Known and Identification of Novel Associations in Biobank-Scale Datasets: A Survey Using UK Biobank and FinnGen

Abstract

Over the last two decades, numerous genome-wide association studies (GWAS) have been performed to unveil the genetic architecture of human complex traits. Despite multiple efforts aimed at the trans-biobank integration of GWAS results, no systematic analysis of the variant-level properties affecting the replication of known associations (or identifying novel ones) in genome-wide meta-analysis has yet been performed using biobank-scale data. To address this issue, we performed a systematic comparison of GWAS summary statistics for 679 complex traits in the UK Biobank (UKB) and FinnGen (FG) cohorts. We identified 37,148 index variants with genome-wide associations with at least one trait in either cohort or in the meta-analysis, only 3528 (9.5%) of which were shared between UKB and FG. Nearly twice as many variants (6577) were replicated in another dataset at the significance level adjusted for the number of variants selected for replication. However, as many as 9230 loci failed to be replicated. Moreover, as many as 5813 loci were observed as significant associations only in meta-analysis results, highlighting the importance of trans-biobank meta-analysis efforts. We showed that variants that failed to replicate in UKB or FG tend to correspond to rare, less pleiotropic variants with lower effect sizes and lower LD score values. Genome-wide associations specific to meta-analysis were also enriched in low-effect variants; however, such variants tended to be more common and have more consistent frequencies between populations. Taken together, our results show a relatively high rate of non-replication of genome-wide associations in the studied cohorts and highlight both widely appreciated and less acknowledged properties of the associations affecting their identification and replication.