Selection of SNP from 50K and 777K arrays to predict breed of origin in cattle

Reliable breed assignment can be performed with SNP. Currently, high density SNP chips are available with large numbers of SNP from which the most informative SNP can be selected for breed assignment. Several methods have been published to select the most informative SNP to distinguish among breeds. In this study, we evaluated Delta, Wright's FST, and Weir and Cockerham's FST, and extended these methods by adding a rule to avoid selection of sets of SNP in high linkage disequilibrium (LD) providing the same information. The SNP that had a r2 value>0.3 with any of the SNP already selected were discarded. The different selection methods were evaluated for both the 50K SNP and 777K Bovine BeadChip. Animals from 4 cattle breeds (989 Holstein Friesian, 97 Groningen White headed, 137 Meuse-Rhine-Yssel, and 64 Dutch Friesian) were genotyped. After editing 30,447 and 452,525 SNP were available for the 50K and 777K SNP chip, respectively. All selection methods showed that only a small set of SNP is needed to differentiate among the 4 Dutch cattle breeds, whereas comparison of the selection methods showed only small differences. In general, the 777K performed marginally better than the 50K BeadChip, especially at higher confidence thresholds. The rule to avoid selection of SNP in high LD reduced the required number of SNP to achieve correct breed assignment. The Global Weir and Cockerham's FST performed marginally better than other selection methods. There was little overlap in the SNP selected from the 2 BeadChips, whereas the number of SNP selected was about the same.