Analysis of crop breeding trials using linear mixed models

Abstract: 

The aim of crop breeding trials is to increase the genetic gain which is typically defined as the increase in genetic performance after one generation in artificial genetic improvement program. This can be reinterpreted as selecting superior varieties from a fixed pool of varieties. For this, we require to reliably predict the genetic performance of the varieties. Linear mixed model is widely used for genomic prediction as it can be constructed to fit the structure of the data.  In particular, the observed trait in crop breeding trials generally exhibit spatial variation and there is a necessity for spatial modelling, which some still neglect in the analysis of crop breeding trials. The most efficient method of analysis is using a one-stage analysis, however, there is wide-spread practice in crop breeding trials to use a two-stage analysis. Typically this involves in the first step, the computation of the predicted variety mean followed by either weighted or unweighted analysis of the variety means in the second step. We present an introductory analysis of crop breeding trial and show in this talk the loss of efficiency involved in employing two-stage analysis in crop breeding trials with simulations based on an analysis of a real wheat breeding trial.