Employing factor analytic tools for selecting high‐performance and stable tropical maize hybrids

Abstract

Genotype‐by‐environment interaction (GEI) is a major concern in tropical maize breeding. Here, we used factor analytic mixed models (FAMM) to study GEI in a tropical maize dataset. Our goal was to select high‐performance and stable hybrids using factor analytic selection tools (FAST) derived from FAMM outputs. Because the dataset comprised two different planting seasons, we also investigated differences in the genetic gains between seasonal and general selection (i.e., for both seasons simultaneously). The trials were installed in a lattice design with 36 hybrids containing two replicates and blocks with six plants each. We evaluated 53 hybrids and seven checks in 48 environments, represented by a combination of locations, years and seasons. We fitted the FAMMs with distinct sets of factors. The best‐fitted FAMM was selected considering a parsimony‐explained variance balance. Using the best FAMM for both seasonal and general analyses, we estimated the following selection tools: overall performance (OP$\textit{OP}$); root mean square deviation (RMSD$\textit{RMSD}$), which represents general stability; and responsiveness , which represents specific stability. We selected hybrids based on an index built with OP$\textit{OP}$ and RMSD$\textit{RMSD}$. Five and three hybrids were selected considering both general and seasonal analyses for the first and second season, respectively. Two hybrids were amongst the top eight for both seasonal analyses (i.e., only first season and only second season simultaenously). Higher genetic gains may be obtained by employing FAMM and using FAST in each season rather than analysing both seasons in a joint fashion.

Analysis of repeated measures data through mixed models: An application in Theobroma grandiflorum breeding

“…To compare the rankings between the first model and the best‐fitted model, we standardized $$OP_i$$ and $$RMSD_i$$ and constructed a selection index, as performed by Chaves et al. (2023), giving equal weights for both. Figure 4 shows the differences between the models' ranks.…”