Chromosome-Specific Genomic Relationships Using Haplotypes for Genomic Prediction and Variance Component Estimation

Current methods for calculating genomic relationships in genomic prediction and variance component estimation typically use SNPs of all autosomes and the resulting genomic relationships on average are remarkably consistent with the expectation of pedigree relationships. However, sib haplotype relationships may have large variations ranging from 0% for sharing opposite parental haplotypes to 100% for sharing identical parental haplotypes due to the small number of crossovers per chromosome (1 crossover per 110Mb per individual in Holstein cattle). Preliminary results showed that chromosome-specific haplotype relationships had considerable variations from the haplotype relationships calculated from all chromosomes. To account for variations of haplotype relationships within each chromosome, we developed mixed model methods for genomic prediction and variance component estimation using chromosome-specific genomic relationships of haplotypes constructed from SNPs of each chromosome. A multifactorial mixed model with haplotype additive and dominance relationships as well as additive and dominance variances of each autosome is used to implement genomic prediction and variance component estimation. Genomic restricted maximum likelihood estimation (GREML) is used to estimate chromosome-specific haplotype variances and the residual variance, and chromosome-specific additive and dominance heritabilities of the haplotypes are calculated based on chromosome-specific variances. Genomic best linear unbiased prediction (GBLUP) is calculated at the last iteration for GREML estimation of variance components.