Genomic prediction in cattle based on sequence data

Summary

One of the factors influencing the accuracy of genomic prediction is the density of SNP data used for prediction. We used sequence genotypes from the 1000 Bull Genomes Project (Run 5) as reference data to impute the whole-genome sequences of around 22,000 Brown Swiss and 15,000 Holstein, Simmental and Swiss Fleckvieh cattle to whole-genome sequences. We used FImpute to obtain HD genotypes and imputed with Minimac from HD to sequence data (16,184,800 variants). We report here the results for non return-rate 56 in heifers which are available already for Brown Swiss. For effect estimation deregressed breeding values of 2,018 Brown Swiss bulls with reliabilities above 0.65 were used. We used the BayesC approach implemented in gbcpp. Accuracy of genomic prediction was calculated as the correlation between the deregressed breeding values and the predicted direct genomic breeding value in a set of 240 of young bulls with accurate breeding values. LD pruned sequence data (5,812,425 SNPs; r=0.412) and sequence data (12,973,772 SNPs; r=0.407) yielded a higher accuracy than 50k data (38,009 SNPs; r=0.400) and missense data (34,184 SNPs; r=0,372).
The results will be further evaluated by investigating more traits and breeds.