Using factor-analytic models and enviromics to select and predict high-performing and stable maize hybrids

• Main Author: Santos, Jenifer Camila Godoy dos
• Publication Date: 2024
• Format: Doctoral thesis
• Language: eng
• Source: Biblioteca Digital de Teses e Dissertações da USP
• Download full: https://www.teses.usp.br/teses/disponiveis/11/11137/tde-10122024-114715/
• topic: Covariáveis ambientais; Environmental covariates; Genotype-by-environment interaction; Interação genótipos por ambientes; Melhoramento preditivo; Modelos multivariados; Multivariate models; Predictive breeding

A major challenge in a maize breeding program is providing clear and meaningful information for genotype selection in the presence of genotype-by-environment interaction (GEI). Since GEI affects phenotypes across different levels of biological organization, environmental factors, such as soil moisture and daily temperature, play a crucial role in shaping quantitative traits. Thus, factor-analytic (FA) linear mixed models and enviromics are some approaches that can assist breeders in selecting and recommending hybrids, uncovering the factors behind GEI, and predicting phenotypes in yet-to-be-seen growing conditions. In this context, the objectives of this study were: (i) to apply an FA model to select high-performing and stable maize hybrids using factor analytic selection tools (FAST); (ii) to develop a special case of FA model considering a separate intercept for each hybrid; (iii) to compare both classes of models and discuss their practical implications for selecting individuals in the context of multi-environment trials (MET); (iv) to employ a robust FA model, which accounts for a blended additive relationship matrix, to predict the hybrids\' performance in non-evaluated environments using different types of environmental covariates; (v) to analyze the model\'s predictive ability by comparing the best observed and predicted values; and (vi) to use thematic maps to analyze performance and stability of these hybrids in any non-evaluated environment. For this, the study relied on pedigree, genomic, and grain yield data from 156 maize hybrids evaluated across 14 tropical environments, as well as on environmental data, including geographical coordinates, weather variables, and soil traits, collected from both the 14 tested and 102,412 untested environments. The results revealed that some tropical environments displayed low genetic correlation, indicating the presence of crossover GEI. Despite this challenge, the FAST approach proved effective in selecting hybrids that not only exhibited high-performance levels but also demonstrated consistency across different environments. The special case of the FA model yielded results that were very similar to those of the traditional FA model. However, the main effects of hybrids, represented by this models intercept, are based on simple averages across environments, and they may not fully capture the complexities introduced by differential environmental conditions. Regarding predictions for non-evaluated environments, the predictive models have proven their effectiveness by delivering accurate results. Ultimately, the thematic maps supplied a visual representation of hybrids\' performance in new locations, and they confirmed that the hybrid with the highest performance and stability, as identified by FAST, would also have good performance and stability in several untested environments. Therefore, this study showcased FA models, when aligned with enviromics, present immense potential in hybrid recommendation and new environment prediction, paving the way for a more efficient and effective future in maize breeding programs.