Marker selection using posterior inclusion probability in genomic prediction models for rice data

Abstract

The objective of this work was to evaluate the performance of the combination of the BayesDπ model for marker selection based on posterior inclusion probability (PIP) and of BayesA in assessing the predictive ability, heritability, and predictive bias for a set of rice phenotypic traits. The Markov chain Monte Carlo algorithm was used for the data analysis. For the calculation of PIP, marker effects were estimated using the BayesDπ method. Subsequently, the ratio between the number of iterations in which each marker had a non-zero effect and the total number of iterations was calculated. The markers were allocated into groups of 2,000, 4,000, 6,000, ..., and 36,901 (entire data set), in descending order of importance. The BayesA method was used to re-estimate the effect of the markers in each group. For comparison purposes, marker effects were also calculated using the BayesA and BayesDπ methods separately. In the proposed model, the PIP proved to be effective in understanding genetic architecture, resulting in a higher predictive ability, as well as in a higher heritability and a lower bias in the selection of the most important markers for genomic prediction compared with the other methods without prior marker selection.