Archived Comments for:
Genome-wide association analysis of thirty one production, health, reproduction and body conformation traits in contemporary U.S. Holstein cows
I have serious concerns about the statistical methods used in this paper. With the recent advances in statistical methods for genome-wide association studies, I am troubled that the authors did not account for pedigree relationships between cows in the study. For an overview of different methods available to correct for cryptic relatedness see Nat Rev Genet. 2010 Jul;11(7):459-63.; J Dairy Sci. 2008 Nov;91(11):4414-23.; and Am J Hum Genet. 2011 Jan 7;88(1):76-82..
Furthermore, I am concerned that the authors used PTA values without deregressing the estimated breeding values, removing parent average effects or correctly weighting to account for heterogeneous variances. See Genet Sel Evol. 2009; 41(1): 55 for a more detailed description. Animals with PTA values with accuracies of zero have no information on their true phenotype; it is simply an average of the parents’ PTA. That is why it is given an accuracy of zero. We can consider the following model of a progeny additive genetic merit: Aprogeny = 1/2Asire + 1/2Adam + phi Phi is the Mendelian sampling term and accounts for the deviation of the gametes received by the progeny from the average parental gamete. It can be shown that in a randomly mating population that the variance of the Mendelian sampling term is equal to half of the additive genetic variance. Although this is not a randomly mating population, the variance of phi is likely to be large. Thus, the true breeding value of an individual can be drastically different than the parent average.
It confuses me that there are more appropriate statistical methods available, but the authors did not employ these methods. The –log10(P-values) seem quite inflated to me, indicative of inappropriate statistical models. I am not confident in any of the results of this manuscript. I am confident that if the PTA values were deregressed, parent average effects were removed, values were weighted according to Garrick et al., and pedigree relationships were accounted for using a genomic relationship matrix, that the genomic regions identified as containing QTLs would be different than the ones reported in this study.
Serious Concerns
18 October 2011
I have serious concerns about the statistical methods used in this paper. With the recent advances in statistical methods for genome-wide association studies, I am troubled that the authors did not account for pedigree relationships between cows in the study. For an overview of different methods available to correct for cryptic relatedness see Nat Rev Genet. 2010 Jul;11(7):459-63.; J Dairy Sci. 2008 Nov;91(11):4414-23.; and Am J Hum Genet. 2011 Jan 7;88(1):76-82..
Furthermore, I am concerned that the authors used PTA values without deregressing the estimated breeding values, removing parent average effects or correctly weighting to account for heterogeneous variances. See Genet Sel Evol. 2009; 41(1): 55 for a more detailed description. Animals with PTA values with accuracies of zero have no information on their true phenotype; it is simply an average of the parents’ PTA. That is why it is given an accuracy of zero. We can consider the following model of a progeny additive genetic merit:
Aprogeny = 1/2Asire + 1/2Adam + phi
Phi is the Mendelian sampling term and accounts for the deviation of the gametes received by the progeny from the average parental gamete. It can be shown that in a randomly mating population that the variance of the Mendelian sampling term is equal to half of the additive genetic variance. Although this is not a randomly mating population, the variance of phi is likely to be large. Thus, the true breeding value of an individual can be drastically different than the parent average.
It confuses me that there are more appropriate statistical methods available, but the authors did not employ these methods. The –log10(P-values) seem quite inflated to me, indicative of inappropriate statistical models. I am not confident in any of the results of this manuscript. I am confident that if the PTA values were deregressed, parent average effects were removed, values were weighted according to Garrick et al., and pedigree relationships were accounted for using a genomic relationship matrix, that the genomic regions identified as containing QTLs would be different than the ones reported in this study.
Competing interests
None declared