Statistical models play a significant role in designing competent breeding programs related to complex traits. Recently; the holo-omics framework has been productively utilized in trait prediction; but it contains many complexities. Therefore; it is desirable to establish prediction accuracy while combining the host's genome and microbiome data. Several methods can be used to combine the two data in the model and study their effectiveness by estimating the prediction accuracy. We validate our holo-omics interaction models with analysis from two publicly available datasets and compare them with genomic and microbiome prediction models. We illustrate that the holo-omics interactive models achieved the highest prediction accuracy in ten out of eleven traits. In particular; the holo-omics interaction matrix estimated using the Hadamard product displayed the highest accuracy in nine out of eleven traits, with the direct holo-omics model and microbiome model showing the highest prediction accuracy in the remaining two traits. We conclude that comparing prediction accuracy in different traits using real data showed important intuitions into the holo-omics architecture of complex traits.
Keywords: breeding program; complex trait; holo-omics; model selection; prediction accuracy; random effect.