Assessing genotype imputation methods for low-coverage sequencing data in populations with differing relatedness and inbreeding levels

Low-coverage sequencing (LCS) followed by genotype imputation has become a cost-efficient approach for obtaining whole-genome SNPs. Several imputation methods for LCS data have been developed over the last decade. However, comparisons of their accuracy in inferring missing genotypes and their effectiveness for downstream analysis such as population genetics have not been comprehensively studied. In the present study, we assessed the imputation performance of five different tools: GLIMPSE2, GeneImp, QUILT2, STITCH and Beagle5.4, using populations simulated by SLiM4 that represent different levels of genetic relatedness and inbreeding. Imputation accuracy was calculated at the level of variant, haplotype and sample. The effectiveness of using imputed genotypes in recovering genetic structure, relatedness, inbreeding coefficients and demographic history was subsequently evaluated. The imputation accuracy of different methods was further tested in a real population of 283 hihi (stitchbird) samples. Our results suggest a high accuracy of all the tested methods on populations with high levels of genetic relatedness. However, in populations with low relatedness, the imputation accuracy differed across different tools and impacted the results of some downstream analyses. The simulation and imputation pipeline presented here can help determine the most suitable imputation method for different population scenarios.