Effects of incorporating the additive relationship on REML estimates of heritability, for grain yield in soybean [Glycine max (L.) Merr.]

摘要

The goal of this research was to analyses the effect of the inclusion of the additive relationship matrix A in the genetic evaluation of progenitors of soybean varieties using mixed linear models. More specifically, we analysed the effects of including A in: i) the restricted maximum likelihood estimation (REML) estimates of the variance components and in ii) the asymptotic variances of REML estimates. Data used in the study were obtained from the national trials network where 39 genotypes were evaluated during 4 years in 15 locations. Estimate of heritability (h~2) by REML was higher using the A matrix (h~2 = 0.15 +- 0.0241) than excluding A from the analysis (h~2 = 0.09 +- 0.0185). Similarly, when A was considered in the model, the coefficient of variation of REML estimate of the additive variance was smaller, compared with the coefficient obtained when the additive relationships were excluded in the analysis (18 to 22). Furthermore, including A in the model reduced the magnitude of the correlation between the REML estimates of the additive variance and the error variances, -0.01 to -0.07, thus improving the accuracy of estimation for both parameters. Finally, breeding values predicted by mixed models resulted in a 21 percent improvement of the Spearman-rank-correlation between the mating rankings predicted by BLUP with the observed rankings at the end of the breeding program, when compared with the same correlation using breeding values predicted by the parents mean value.