False discovery rate control in genome-wide association studies with population structure

Abstract

We present a comprehensive statistical framework to analyze data from genome-wide association studies of polygenic traits, producing interpretable findings while controlling the false discovery rate. In contrast with standard approaches, our method can leverage sophisticated multivariate algorithms but makes no parametric assumptions about the unknown relation between genotypes and phenotype. Instead, we recognize that genotypes can be considered as a random sample from an appropriate model, encapsulating our knowledge of genetic inheritance and human populations. This allows the generation of imperfect copies (knockoffs) of these variables that serve as ideal negative controls, correcting for linkage disequilibrium and accounting for unknown population structure, which may be due to diverse ancestries or familial relatedness. The validity and effectiveness of our method are demonstrated by extensive simulations and by applications to the UK Biobank data. These analyses confirm our method is powerful relative to state-of-the-art alternatives, while comparisons with other studies validate most of our discoveries. Finally, fast software is made available for researchers to analyze Biobank-scale datasets.

The Association Between Genomic Heterozygosity and Carcass Merit in Cattle

“…In addition, many SNPs detected to have associations with the traits of interest to the present study, namely those detected to have additive associations with CC and, to a lesser extent, additive associations with CF, were not present within clear peaks or signals. This has been previously attributed to numerous factors such as allele frequency, linkage disequilibrium and population structure, among others factors ( Tabangin et al, 2009 ; Platt et al, 2010 ; Sesia et al, 2021 ). Finally, there was some overlap between the detected additive and dominance QTL regions in the present study; the presence of overlapping additive and dominance QTL regions have been detected in previous cattle studies ( Kim et al, 2003 ; Powell et al, 2013 ; Li et al, 2017 ).…”

Individualized conditional independence testing under model-X with heterogeneous samples and interactions

“…, X 5 ] = 1 otherwise. In principle, it would also be possible to generate synthetic CRT treatments and knockoffs without the above independence assumption, as in [16], but that would require ad-hoc techniques which we do not discuss here for lack of space. In any case, the independence assumption should have little impact on the practical validity of our tests given that the sample size is quite large.…”

“…Further, the abundance of variables and observations in those studies will translate into further advantages for our methodology, which generally tends to perform better when applied to larger-scale data sets. Of course, computing powerful test statistics based on extremely big data sets raises additional computational challenges, but the model-X framework has already been demonstrated to be quite scalable [16,48,52,74].…”

“…, where P Z is irrelevant because Z will be treated as fixed, P X|Z is assumed to be known, and P Y |X,Z is the target of inference, which may be arbitrary and completely unknown. Finally, we imagine different individuals are independent of one another, but this simplifying assumption could be relaxed at the cost of more involved calculations if the sample dependencies are known [16].…”

“…The ability to discover such individualized associations would be especially useful when dealing with heterogeneous populations in which subsets of individuals have disparate features and may respond incongruously to a treatment. For example, people with diverse ancestries do not all have the same allele frequencies, and even if the likelihood of the phenotype given all relevant variables is assumed to be universal [16], interactions can result in some variants having large effects for some groups while being practically useless for others.…”