DIMENSION REDUCTION IN SELECTIVE GENOTYPING MODEL FOR GENOMIC SELECTION ON FEED EFFICIENCY TRAITS IN NELLORE CATTLE

Abstract

Recent advances in technologies for genotyping individual animals, most notably the development of high throughput assays for dense genotyping of SNP, have led the possibility of increasing the genetic gain on a beef cattle population, mainly on traits that are hard and costly to measure as feed efficiency, for example. However, the large number of obtained information from genotyping animals has generated controversy among researchers about which is the best way to evaluate such information and if the existing methods currently applied are capable of reproduce variability in a reliable way of the polygenic traits. This concern has become even greater when the phenotypes that are used in such methods are derived from a small number of animals, the considered "elite" animals. The major challenge then is to predict the genetic value of animals by building statistical models that can handle a database that possesses high number of variables with a small number of observations. Several regression models have been used where the marker effect is treated as a random variable, however, the inter-correlation, due to linkage disequilibrium between markers, causes multicollinearity which significantly increases the standard error. Alternatively, models of dimension reduction have been proposed to deal with such effects. Thus the objective of this project will be test the methodology of supervised principal components in order to reduce the dimensionality in genomic selection of feed efficiency traits in Nellore cattle, where a pre-selection of SNPs is performed based on the relationship between the SNP and the phenotype.