A reassessment of Hardy-Weinberg equilibrium filtering in large sample Genomic studies.

Abstract

Hardy Weinberg Equilibrium (HWE) is a fundamental principle of population genetics. Adherence to HWE, using a p-value filter, is used as a quality control measure to remove potential genotyping errors prior to certain analyses. Larger sample sizes increase power to differentiate smaller effect sizes, but will also affect methods of quality control. Here, we test the effects of current methods of HWE QC filtering on varying sample sizes up to 486,178 subjects for imputed and Whole Exome Sequencing (WES) genotypes using data from the UK Biobank and propose potential alternative filtering methods. METHODSSimulations were performed on imputed genotype data using chromosome 1. WES GWAS (Genome Wide Association Study) was performed using PLINK2. RESULTSOur simulations on the imputed data from Chromosome 1 show a progressive increase in the number of SNPs eliminated from analysis as sample sizes increase. As the HWE p-value filter remains constant at p<1e-15, the number of SNPs removed increases from 1.66% at n=10,000 to 18.86% at n=486,178 in a multi-ancestry cohort and from 0.002% at n=10,000 to 0.334% at n=300,000 in a European ancestry cohort. Greater reductions are shown in WES analysis with a 11.91% reduction in analyzed SNPs in a European ancestry cohort n=362,192, and a 32.70% reduction in SNPs in a multi-ancestry dataset n=463,605. Using a sample size specific HWE p-value cutoff removes [~] 2.25% of SNPs in the all ancestry cohort across all sample sizes, but does not currently scale beyond 300,000 samples. A hard cutoff of +/- 20% deviation from HWE produces the most consistent results and scales across all sample sizes but requires additional user steps. CONCLUSIONTesting for deviance from HWE may still be an important quality control step in GWAS studies, however we demonstrate here that using an HWE p-value threshold that is acceptable for smaller sample sizes will be inappropriate for large sample studies due to an unnecessarily high number of variants removed prior to analysis. Rather than exclude variants that fail HWE prior to analysis it may be better to include all variants in the analysis and examine their deviation from HWE afterward. We believe that adjusting the cutoffs will be even more important for large whole genome sequencing results and more diverse population studies. KEY TAKEAWAYSO_LICurrent thresholds for assessing HWE are impractical for large sample sizes. C_LIO_LIFiltering imputed datasets for HWE regardless of sample size is unnecessary and in fact detrimental if you have a diverse, mixed, or unknown ancestry cohort. C_LIO_LIWES data shows more distributed deviation from HWE for all Minor Allele Frequencies (MAF). C_LIO_LIWe present an alternative p-value filter for HWE for large sample sizes. C_LIO_LIWe recommend that all genotype data (imputed, WES or WGS) should be analyzed, HWE computed, results combined, and then filtered post-hoc. C_LI