On the Identifiability of Genetic Parameters for Growth in Mice With a Massively Deep Pedigree.

Abstract

In models with direct and maternal genetic effects, structural features of the data are a potential source of bias and low accuracy of estimates for genetic covariance parameters. One of the well-known reasons for such poor practical identifiability is the lack of dams with own observations. So far, however, no attention has been paid to the impact close relationships may have. Therefore, this genetic-statistical analysis of growth traits in two unselected mouse lines includes investigations on practical identifiability of genetic (co-)variances in the light of the observed high levels of co-ancestry, resulting from massively deep pedigrees. Body weight data had been collected over 33 years (from 1978 to 2011; 145 and 118 generations per line), amounting to approximately 115,000 observations in total for body weight at three developmental stages. Additional analyses of simulated data using the original pedigree structure of one line provided insight into the bias and precision of estimates. Further, closeness to pair-wise structural non-identifiability of genetic (co-)variances was quantified. In univariate analyses, we found genetic correlations between direct and maternal effects all positive for body mass traits at different ages up to mating, except for a single small negative estimate. Overall, multivariate analyses returned somewhat stronger correlations, whereby signs remained unchanged. Simulations showed a tendency toward an upward bias of the direct-maternal genetic correlations and other parameters, especially when the true correlations were higher. For all traits indicators for structural non-identifiability were narrowly close (> 0.998) to unity, the point at which a pair of covariance components no longer can be identified. This narrowness was stronger for separate partitions of data from later generations with higher average inbreeding and within-generation co-ancestry. In conclusion, in models with direct and maternal genetic effects, strong co-ancestry between parents is another feature of the data structure that may result in bias and inflated standard errors of estimated genetic parameters.