Genomic estimates of Identity-By-Descent relationships in large scale data sets

Background Genomic relationship and inbreeding estimates are either based on genetic drift (e.g. the Genomic Relationship Matrix; GRM ), homozygosity (e.g. Runs of Homozygosity; ROH), or Identity-By-Descent (IBD). A genomic IBD-based relationship matrix, G la , is obtained by linkage analysis which uses genomic data to distinguish paternal versus maternal inheritances of chromosomal segments to replace the 50/50 probabilities used to calculate pedigree-based relationships ( A matrix). Our aim was to develop a fast approximate algorithm, FGla, to estimate the G la matrix in large complex pedigrees making use of dense marker genotypes, and to compare G la to A , GRM and ROH based inbreeding ( F ROH ) in simulated and a large scale Norwegian Red Cattle (NRF) data set. Results Given pedigree data and ≥ 3 generations of 45 k marker genotypes, marker positions were detected that unambiguously identified maternal/paternal inheritance, and inheritances at intermediate positions were imputed by the Viterbi algorithm from the positions with known inheritance. Any remaining unknown inheritances were randomly sampled (paternal or maternal), and the sampling errors that this introduced were averaged out by the large number of marker loci used (correlation between replicated estimates: 0.9998). Also, calculations were limited to the relationship coefficients that were actually needed, assuming that relationships for a limited set of candidates were needed. The accuracy of estimated G la coefficients increased from 0.971 to 0.998, when genotyping increased from the actually genotyped NRF cattle towards all pedigreed animals. The accuracy of the GRM was 0.936, but required only genotyping of the animals whose relationships were needed. G la relationships were approximately unbiased in the Best Linear Unbiased Prediction (BLUP) sense. Hence, if G la based inbreeding management predicts an increase in relationships then an identical increase in true IBD relationships is expected. G la uses the same base population as A , namely that of the pedigree. Conclusions An approximate computationally efficient multipoint linkage analysis algorithm was developed to estimate unbiased IBD-based relationship and inbreeding coefficients. Its unbiasedness and precise definition of the base population makes it well suited for the genomic management of inbreeding and genomic optimal contribution selection. In addition, G la based optimal contribution selection is neutral with respect to allele frequency changes.