Is it possible to skip the SNP selection step for breed assignment?

Breed assignment of animals generally relies on three main steps: 1) Selection of a SNP panel that allows to distinguish between the breeds of interest, 2) Training of a classification algorithm, and 3) Validation of this classification on new animals. In recent years, to select the most informative SNPs, more complex methodologies have been developed, e.g., by the combination of several methods. There is no consensus about the protocol to follow to select the best SNPs and many questions stay unanswered. How to select the most informative SNPs? What is the optimal number of SNPs to select? To solve these issues, we proposed to skip the SNP selection step by using a genomic relationship matrix based on all available SNPs to assign animals to their breed of origin. Three new methodologies were developed for three cattle breeds: 1) Breed assignment based on the highest mean relatedness of an animal to the reference population of each of the three breeds, 2) Breed assignment based on the highest standard deviation (SD) of the relatedness of an animal to the reference population of each of the three breeds, and 3) A methodology that combines the values of mean and SD of the relatedness of each animal in a linear support vector machine model (SVM). These new methodologies were compared with a control methodology: a previously developed model based on a reduced SNP panel. The linear SVM achieved a similar percentage of correct assignment as the control methodology, and was substantially faster to compute. The main advantage of using the new methodology, based on the linear SVM, is to bypass the SNP selection step.