Principal genetic components analysis for carcass and meat quality traits in Nellore cattle

Principal component analysis (PCA) is widely used in data processing aiming dimensionality reduction. The goal of this study was to compare a standard multivariate model with reduced rank models for estimating genetic parameters of carcass and meat quality (MQ) traits in Nellore cattle. A total of 3,229; 3,332;3,219;2,433 3,226; 3,233 and 24,962 Nellore records for hot carcass weight (HCW), 12-13^th rib eye area (REA), backfat thickness (BF), lipid content (LC), marbling score (MS),Warner-Braztler Shear Force (WBSF) and weight at 550 days (W550), respectively, were used . A standard multivariate animal model (MV) and reduced-rank models fitting two (PC2), three (PC3), and four (PC4) genetic principal components, were applied. Models were compared by Akaike Information Criterion (AIC) and Schawrz's Bayesian Information Criterion (BIC). Direct genetic random effects and fixed effects of contemporary group (farm and year of birth, and management group at yearling) and age of animal at recording as covariate (linear) were considered in the models for all traits. The residual (co)variance matrices were assumed to have full rank. According to the results of AIC, the model fitting four principal components (PC4) provided the best fit, while BIC results pointed out that two principal components (PC2) were enough to model the genetic variance. Compared to PC2, however, the genetic (co)variances obtained with PC4 were closer to those obtained with MV. The heritability estimates ranged from 0.04 (LC) to 0.28 (REA) with MV and PC4, and from 0.01 (LC) to 0.22 (REA) with PC2. In general, four main components wererequired to model the structure of genetic (co)variances for carcass and meat quality traits. Additionally, the reduced rank model decreased the number of parameters to be estimated for the genetic (co)variance matrix from 28 to 22, without reducing the quality of adjustment.