Estimation of heritability with genomic information by method R

Area of Science:

• Quantitative Genetics
• Animal Breeding
• Genomic Selection

Background:

• Estimating heritability using large genomic models with methods like restricted maximum likelihood (REML) or Bayesian Gibbs sampling is computationally intensive.
• Alternative methods, such as Method R and Maximum Predictivity (MaxPred), offer lower computational costs for heritability estimation.

Purpose of the Study:

• To compare the accuracy and computational efficiency of heritability estimation using Average Information REML (AI-REML), Method R, and MaxPred in large genomic models.
• To evaluate the performance of Method R and MaxPred for estimating heritability with genomic information across multiple generations.

Main Methods:

• A simulated population with ten generations of 5000 animals each was used, featuring a trait with a heritability of 0.3 and 50k SNPs per animal.
• Heritability was estimated using AI-REML (with and without genomics) and Method R (with genomics) within a Genomic Best Linear Unbiased Prediction (GBLUP) framework.
• Method R estimated heritability by ensuring the regression coefficient of GEBV from whole data on GEBV from partial data (last generation's phenotypes removed) equals one.
• MaxPred estimated heritability by maximizing the correlation between adjusted phenotypes and GEBV from partial data.

Main Results:

• Method R with genomics reduced computation time by 83% (from 9.5 to 1.6 hours) compared to AI-REML with genomics, while yielding a similar heritability estimate (0.30 ± 0.04 vs. 0.26 ± 0.01).
• MaxPred showed that predictivity increased with more data but differences between subsets were minimal (≤0.01), suggesting it's not a reliable indicator for heritability estimation.
• AI-REML without genomics estimated heritability at 0.30 ± 0.01, comparable to Method R with genomics.

Conclusions:

• Method R is a computationally efficient alternative for estimating heritability in large genomic datasets, especially when multiple generations of data are available.
• While Method R offers significant computational savings, its standard error can be high with limited iterations.
• MaxPred is not recommended as a primary method for heritability estimation due to its low sensitivity to varying data subsets and heritabilities.