Data from: The impact of variable degrees of freedom and scale parameters in Bayesian methods for genomic prediction in Chinese Simmental beef cattle

Three conventional Bayesian approaches (BayesA, BayesB and BayesCπ) have
been demonstrated to be powerful in predicting genomic merit for complex
traits in livestock. A priori, these Bayesian models assume that the
non-zero SNP effects (marginally) follow a t-distribution depending on two
fixed hyperparameters, degrees of freedom and scale parameters. In this
study, we performed genomic prediction in Chinese Simmental beef cattle
and treated degrees of freedom and scale parameters as unknown with
inappropriate priors. Furthermore, we compared the modified methods
(BayesFA, BayesFB and BayesFCπ) with their corresponding counterparts
using simulation datasets. We found that the modified methods with
distribution assumed to the two hyperparameters were beneficial for
improving the predictive accuracy. Our results showed that the predictive
accuracies of the modified methods were slightly higher than those of
their counterparts especially for traits with low heritability and a small
number of QTLs. Moreover, cross-validation analysis for three traits,
namely carcass weight, live weight and tenderloin weight, in 1136
Simmental beef cattle suggested that predictive accuracy of BayesFCπ
noticeably outperformed BayesCπ with the highest increase (3.8%) for live
weight using the cohort masking cross-validation.