Simulation Study on Multi-Trait Genomic Prediction Using Eigen Decomposition of Genomic Relationship Matrix: A Case...

The performance of multi-trait genomic prediction was assessed by simulating phenotypic data with the publicly available information on the genomic relationship matrix for 9850 Japanese Black cattle utilizing eigen decomposition. Variance component estimation and genomic breeding value prediction were performed by analyzing simulated phenotypic data of two different traits with four simulation settings of varied heritability (0.1, 0.4, and 0.7), genetic correlation (0.6 and 0.95), and phenotypic correlation (0.4 and 0.8). A total of 1000, 3000, and 5000 animals were selected as training populations. The performance of the algorithm for proven and young was also examined using simulated and real phenotypic data. Core animals were chosen from the 9850 animals to minimize the genetic variance of non-core animals conditional on the core animals using the genomic relationship matrix. Five sets of core animals were provided, and those explaining 25%, 50%, 75%, 90%, and 99% of the genetic variance were first explained. The multi-trait model gave greater prediction accuracy than the single-trait model for both traits, and not only heritability and genetic correlation but phenotypic correlation affected the accuracy in the current settings. Core animals explaining 90% and 99% variance could yield results similar to those obtained by using the original genomic relationship matrix.