Comparing estimates of genetic variance across different relationship models

Area of Science:

• Quantitative genetics
• Statistical genetics
• Population genetics

Background:

• Mixed models are standard for estimating genetic variances and heritabilities using relationship information.
• Different relationship models yield varying estimates due to differing base populations.
• Direct comparison of these estimates is problematic.

Purpose of the Study:

• To present a novel method for comparing variance component estimates across diverse relationship models.
• To enable accurate comparisons of genetic variances and heritabilities.
• To address the issue of differing implied base populations.

Main Methods:

• Propose a method to reference genetic variances to a common population.
• Introduce a statistic, Dk (average self-relationship minus average relationship), for scaling estimates.
• Utilize mixed models and relationship matrices for variance component estimation.

Main Results:

• The proposed method scales variance estimates to a comparable base population.
• Dk values close to 1 are typical for standard relationship models.
• Deep pedigrees, identity-by-state, and non-parametric kernels can overestimate genetic variance and heritability.
• Mice data confirmed overestimation of heritability for identity-by-state and kernel methods.

Conclusions:

• The Dk-weighting method allows for accurate comparisons of genetic variances and heritabilities across different relationship models.
• This approach helps prevent erroneous conclusions, such as the "missing heritability" problem.
• Standardizing estimates to a common reference population is crucial for robust genetic analyses.