Accurate imputation of inversions in human genomes using different algorithms and data sources

Complex genomic regions harbor different structural arrangements that can mutate quite rapidly, which makes determining their functional effects very difficult. Characterization of inversions originated by homologous mechanisms is especially challenging due to the presence of inverted repeats at the breakpoints and the fact that most of them are recurrent. Imputation is a useful method to infer missing genotypes, but it has been mainly limited to simple variants and little is known about how well it works for human inversions. Here, we tested five common imputation programs to impute a set of 52 inversions, which have been experimentally genotyped in multiple samples and lacked perfectly linked single nucleotide polymorphisms (SNPs). Using whole-genome sequencing data and simulated microarrays with variable SNP density, we found that 40.4%-75.5% of inversions could be accurately imputed in three human populations by at least one program, with results depending mostly on inversion recurrence and the number of available SNPs and genotyped samples. Besides, genotype probability filtering was a key factor for inversion imputation accuracy. In particular, Minimac4 and IMPUTE5 showed more accurately imputed inversions and less poorly imputed individuals with respect to the other methods. This work therefore contributes to optimizing inversion imputation in order to study their functional impact.