Effect of reduced genomic representation on using runs of homozygosity for inbreeding characterization

Abstract

Runs of homozygosity (ROHs) are proxy for genomic Identical-by-Descent segments and are increasingly used to measure individual inbreeding. ROHs analyses are mostly carried out on SNPs-arrays and whole-genome-sequencing data. Softwares recurrently used for their detection usually assume that genomic positions which have not been genotyped are non-variant. This might be true for whole-genome-sequencing data, but not for reduced genomic representations and can lead to spurious ROHs detection. We simulated the outputs of whole-genome-sequencing, two SNP-arrays and RAD-sequencing for three populations with different sizes. We compare the results of ROHs calling with two softwares: PLINK and RZooRoH. We demonstrate that to obtain meaningful estimates of inbreeding coefficients, RZooRoH requires fraction of genome seven times smaller compared to PLINK. When the SNP density is above 20 SNPs/Mb for PLINK and 3 SNPs/Mb for RZooRoH, ranks of ROHs-based inbreeding coefficients are conserved among individuals. With reduced genomic representations, ROHs distributions are consistently biased towards an underestimation of the total numbers of small and an overestimation of the total numbers of large ROHs, except for RZooRoH and high-density SNPs-arrays. We conclude that both ROHs-based inbreeding coefficients and ROHs distributions exact quantification are highly dependent on the fraction of genome sequenced and should thus be treated with caution. However, relative inbreeding estimates, such as comparison between individuals or populations, are reliable with reduced genomic representations providing that the fraction of genome sequenced is large enough. Consequently, we advise researchers working with reduced genomic data to use SNPs-independent measures or model-based ROHs calling methods for inbreeding estimations.

Purging of highly deleterious alleles through an extreme bottleneck

“…We used RZooRoH (Bertrand et al, 2019) to detect ROH in our sampled individuals although we acknowledge that the use of reduced representation and lcWGS for ROH detection is still a fraught process (Duntsch et al, 2021;Lavanchy and Goudet, 2022). Despite diZerences in absolute estimates of FROH there is strong correlation between methods and data types (Duntsch et al, 2021) suggesting that relative comparisons are eZective even with sparse data.…”

“…Across samples, we expect there to be little bias introduced by diZerences in mean depth of coverage, as the wild individuals have the lowest mean depth (Supplementary Table 1) but the greatest number of short ROH (Figure 2bc). With incomplete representation of the genome, ROH detection methods tend to underestimate the number of short ROH (Lavanchy and Goudet, 2022), therefore the dropout of sites due to low depth in the wild individuals has not substantially biased the estimation of relative FROH and ROH tract lengths. The estimation of FROH with such approaches is less sensitive to data quality than is the determination of the precise coordinates of ROH (Duntsch et al, 2021;Lavanchy and Goudet, 2022).…”

“…With incomplete representation of the genome, ROH detection methods tend to underestimate the number of short ROH (Lavanchy and Goudet, 2022), therefore the dropout of sites due to low depth in the wild individuals has not substantially biased the estimation of relative FROH and ROH tract lengths. The estimation of FROH with such approaches is less sensitive to data quality than is the determination of the precise coordinates of ROH (Duntsch et al, 2021;Lavanchy and Goudet, 2022). Apart from the wild male, all samples included in our analysis had a mean per-base coverage greater than 2, and so we expect most of the genome to have been sampled, although minimally for many sites.…”

Detecting inbreeding depression in structured populations

“…One advantage of these methods is that they do not depend on allelic frequencies which can be very valuable when only a few individuals are available. However, it has been shown that these coefficients, and especially F ROH , are sensitive to SNP density and parameters used, and there is no consensus on what is the most suitable set of parameters at present [23,18]. How to quantify ID, although central to conservation genetics for decades [16], is still debated.…”