Imputation of Missing Genotypes From Sparse to High Density Using Long-Range Phasing

Related individuals share potentially long chromosome segments that trace to a common ancestor. We describe a phasing algorithm (ChromoPhase) that utilizes this characteristic of finite populations to phase large sections of a chromosome. In addition to phasing, our method imputes missing genotypes in individuals genotyped at lower marker density when more densely genotyped relatives are available. ChromoPhase uses a pedigree to collect an individual's (the proband) surrogate parents and offspring and uses genotypic similarity to identify its genomic surrogates. The algorithm then cycles through the relatives and genomic surrogates one at a time to find shared chromosome segments. Once a segment has been identified, any missing information in the proband is filled in with information from the relative. We tested ChromoPhase in a simulated population consisting of 400 individuals at a marker density of 1500/M, which is approximately equivalent to a 50K bovine single nucleotide polymorphism chip. In simulated data, 99.9% loci were correctly phased and, when imputing from 100 to 1500 markers, more than 87% of missing genotypes were correctly imputed. Performance increased when the number of generations available in the pedigree increased, but was reduced when the sparse genotype contained fewer loci. However, in simulated data, ChromoPhase correctly imputed at least 12% more genotypes than fastPHASE, depending on sparse marker density. We also tested the algorithm in a real Holstein cattle data set to impute 50K genotypes in animals with a sparse 3K genotype. In these data 92% of genotypes were correctly imputed in animals with a genotyped sire. We evaluated the accuracy of genomic predictions with the dense, sparse, and imputed simulated data sets and show that the reduction in genomic evaluation accuracy is modest even with imperfectly imputed genotype data. Our results demonstrate that imputation of missing genotypes, and potentially full genome sequence, using long-range phasing is feasible