Evaluation of genomic selection methods for predicting fiber quality traits in Upland cotton.

Abstract

The use of genomic selection (GS) has stimulated a new way to utilize molecular markers in breeding for complex traits in the absence of phenotypic data. GS can potentially decrease breeding cycle by selecting the progeny in the early stages. The objective of this study was to experimentally evaluate the potential value of genomic selection in Upland cotton breeding. Six fiber quality traits were obtained in 3years of replicated field trials in Starkville, MS. Genotyping-by-sequencing-based genotyping was performed using 550 recombinant inbred lines of the multi-parent advanced generation inter-cross population, and 6292 molecular markers were used for the GS analysis. Several methods were compared including genomic BLUP (GBLUP), ridge regression BLUP (rrBLUP), BayesB, Bayesian LASSO, and reproducing kernel hilbert spaces (RKHS). The average heritability (h2) ranged from 0.38 to 0.88 for all tested traits across the 3years evaluated. BayesB predicted the highest accuracies among the five GS methods tested. The prediction ability (PA) and prediction accuracy (PACC) varied widely across 3years for all tested traits and the highest PA and PACC were 0.65, and 0.69, respectively, in 2010 for fiber elongation. Marker density and training population size appeared to be very important factors for PA and PACC in GS. Results indicated that BayesB-based GS method could predict genomic estimated breeding value efficiently in Upland cotton fiber quality attributes and has great potential utility in breeding by reducing cost and time.