Assessing Hardy-Weinberg Equilibrium in T2T-aligned 1000 Genomes Project

Abstract

Quality control of markers in genome-wide association studies often includes testing for Hardy-Weinberg equilibrium (HWE). However, this is usually implemented in a homogeneous population without stratifying by sex. Previous work indicates sex-based selection at numerous autosomal loci in cohorts with active recruitment. Sex chromosome sequences can also interfere with autosomal SNPs. We examined genome-wide sex-specific HWE deviations across populations in the telomere-to-telomere (T2Tv2)-aligned high-coverage whole genome sequence of the 1000 Genomes Project data of 2,490 individuals. Our analysis was restricted to bi-allelic SNPs with non-missing genotypes and MAF>=5% in both sexes of the five super-populations. We employed a robust allele-based approach for HWE testing, which enabled the quantification of directional deviations from HWE. A second-order omnibus meta-analysis combining results from the five super-populations and both sexes revealed that 0.9% autosomal SNPs exhibited a significant deviation from HWE at p<5e-8. Most of these deviations were found to be associated with genomic features relating to poor sequence quality. Filtering results to reliable genomic regions yielded 255 autosomal and 1 NPR X chromosomal SNPs, of which 140 autosomal SNPs also showed significant heterogeneity across populations but not across sexes. 8 SNPs in a 15-bp region on chr14 showed excess heterozygosity in both sexes of the AFR (African) super-population. We also generated a well-performing multivariate predictor of HWD (deviation from HWE) using multiple sequence features, which could be combined with HWD estimates in future studies to select SNPs that deviate from HWE due to technical rather than biological reasons. Author SummaryWe conducted a specific quality control test, which compares the observed and expected genotype counts, on an updated version of the 1000 Genomes Project whole genome sequence data generated on [~]2500 individuals. We first performed this analysis by grouping the data by ancestry and sex. We then combined and contrasted the group results. We found that most regions that differed between observed and expected counts overlapped regions of the genome which are difficult to sequence using current short read technology. In the remaining regions we found an interesting cluster of SNPs in a single ancestry, where there is a gross excess of heterozygous genotypes. GWASes typically use a standard strict threshold for this quality control test for genotyping arrays to remove SNPs. Here we suggest a more nuanced approach that is applicable to whole genome sequence data.