META-ANALYSIS FOR EVALUATING THE EFFICIENCY OF GENOMIC SELECTION IN CEREALS

Abstract

Genomic selection (GS) is used to predict the merit of a genotype with respect to a quantitative trait from molecular or genomic data. Statistically, GS requires fitting a regression model with multiple predictors associated with the molecular markers (MM) states. The model is calibrated in a population with phenotypic and genomic data. The abundance and correlation of MM information make model estimation challenging. For that reason there are diverse strategies to adjust the model: based on best linear unbiased predictors (BLUP), Bayesian regressions and machine learning methods. The correlation between the observed phenotype and the predicted genetic merit by the fitted model provides a measure of the efficiency (predictive ability) of the GS. The objective of this work was to perform a metaanalysis on the efficiency of GS in cereals. A systematic review of related GS studies and a meta-analysis, in wheat and maize, was carried out to obtain a global measure of GS efficiency under different scenarios (MM quantity and statistical models used in GS). The meta-analysis indicated an average correlation coefficient of 0.61 between observed and predicted genetic merits. There were no significant differences in the efficiency of the GS based on BLUP (RR-BLUP and GBLUP), the most common statistical approach. The increase of MM data, make GS efficiency do not vary widely.

Keywords

• systematic review
• random effects model
• forest plot
• predictive accuracy