Weighted likelihood inference of genomic autozygosity patterns in dense genotype data

Abstract

BackgroundGenomic regions of autozygosity (ROA) arise when an individual is homozygous for haplotypes inherited identical-by-descent from ancestors shared by both parents. Over the past decade, they have gained importance for understanding evolutionary history and the genetic basis of complex diseases and traits. However, methods to infer ROA in dense genotype data have not evolved in step with advances in genome technology that now enable us to rapidly create large high-resolution genotype datasets, limiting our ability to investigate their constituent ROA patterns.MethodsWe report a weighted likelihood approach for inferring ROA in dense genotype data that accounts for autocorrelation among genotyped positions and the possibilities of unobserved mutation and recombination events, and variability in the confidence of individual genotype calls in whole genome sequence (WGS) data.ResultsForward-time genetic simulations under two demographic scenarios that reflect situations where inbreeding and its effect on fitness are of interest suggest this approach is better powered than existing state-of-the-art methods to infer ROA at marker densities consistent with WGS and popular microarray genotyping platforms used in human and non-human studies. Moreover, we present evidence that suggests this approach is able to distinguish ROA arising via consanguinity from ROA arising via endogamy. Using subsets of The 1000 Genomes Project Phase 3 data we show that, relative to WGS, intermediate and long ROA are captured robustly with popular microarray platforms, while detection of short ROA is more variable and improves with marker density. Worldwide ROA patterns inferred from WGS data are found to accord well with those previously reported on the basis of microarray genotype data. Finally, we highlight the potential of this approach to detect genomic regions enriched for autozygosity signals in one group relative to another based upon comparisons of per-individual autozygosity likelihoods instead of inferred ROA frequencies.ConclusionsThis weighted likelihood ROA inference approach can assist population- and disease-geneticists working with a wide variety of data types and species to explore ROA patterns and to identify genomic regions with differential ROA signals among groups, thereby advancing our understanding of evolutionary history and the role of recessive variation in phenotypic variation and disease.Electronic supplementary materialThe online version of this article (doi:10.1186/s12864-017-4312-3) contains supplementary material, which is available to authorized users.

Detection and classification of hard and soft sweeps from unphased genotypes by multilocus genotype identity

“…Moreover, the proportion of individuals with runs of homozygosity of intermediate length is negatively correlated (correlation coefficient = −0.26, p -value = 1.02 < 10 −3 ) with log 10 (BF) (Table S2), likely due to the narrower genomic signature left behind by soft sweeps relative to hard sweeps. In contrast, we observe no significant correlation for smaller runs of homozygosity (classes 2 and 3), which have also been proposed to potentially be affected by selective sweeps [Pemberton et al, 2012, Blant et al, 2017].…”

“…We further explored the existence of a more general relationship between top sweep candidates and the prevalence and length of runs of homozygosity. Previous research has indicated that short-to-intermediate runs of homozygosity spanning tens to hundreds of kilobases are characteristic of recent sweeps [Pemberton et al, 2012, Blant et al, 2017], and we sought to examine whether there was a correlation of G123 or sweep softness (using log 10 (BF) as proxy) with the proportion of individuals falling in a run of homozygosity of specific length. To this end, we intersected our top candidates lists with the inferred coordinates of short to intermediate runs of homozygosity from Blant et al [2017].…”

“…Previous research has indicated that short-to-intermediate runs of homozygosity spanning tens to hundreds of kilobases are characteristic of recent sweeps [Pemberton et al, 2012, Blant et al, 2017], and we sought to examine whether there was a correlation of G123 or sweep softness (using log 10 (BF) as proxy) with the proportion of individuals falling in a run of homozygosity of specific length. To this end, we intersected our top candidates lists with the inferred coordinates of short to intermediate runs of homozygosity from Blant et al [2017]. We found that the proportion of individuals with runs of homozygosity of intermediate length (class 4) is positively correlated (correlation coefficient = 0.32, p -value = 3.66 × 10 −5 ) with G123 (Table S2), likely due to stronger and more recent sweeps generating larger G123.…”

Detection and Classification of Hard and Soft Sweeps from Unphased Genotypes by Multilocus Genotype Identity

“…Moreover, the proportion of individuals with runs of homozygosity of intermediate length is negatively correlated (correlation coefficient = 20.26, P-value = 1:02 3 10 23 ) with log 10 ðBFÞ (Table S2), likely due to the narrower genomic signature left behind by soft sweeps relative to hard sweeps. In contrast, we observe no significant correlation for smaller runs of homozygosity (classes 2 and 3), which have also been proposed to potentially be affected by selective sweeps (Pemberton et al 2012;Blant et al 2017).…”

“…Previous research has indicated that short-to-intermediate runs of homozygosity spanning tens to hundreds of kilobases are characteristic of recent sweeps (Pemberton et al 2012;Blant et al 2017), and we sought to examine whether there was a correlation of G123 or sweep softness [using log 10 ðBFÞ as proxy] with the proportion of individuals falling in a run of homozygosity of specific length. To this end, we intersected our top candidates lists with the inferred coordinates of short to intermediate runs of homozygosity from Blant et al (2017). We found that the proportion of individuals with runs of homozygosity of intermediate length (class 4) is positively correlated (correlation coefficient = 0.32, P-value = 3:66 3 10 25 ) with G123 (Table S2), likely due to stronger and more recent sweeps generating larger G123.…”