Envirome-Wide Associations Enhance Multi-Year Genome-Based Prediction of Historical Wheat Breeding Data

Abstract

Linking high-throughput environmental data (enviromics) into genomic prediction (GP) is a cost-effective strategy for increasing selection intensity under genotype-by-environment interactions (GxE). This study developed a data-driven approach based on Environment-Phenotype Associations (EPA) aimed at recycling important GxE information from historical breeding data. EPA was developed in two applications: (1) scanning a secondary source of genetic variation, weighted from the shared reaction norms of past-evaluated genotypes; (2) pinpointing weights of the similarity among trial-sites (locations), given the historical impact of each envirotyping data variable for a given site. Then, the EPA outcomes were integrated into multi-environment GP models through a new single-step GBLUP. The wheat trial data used included 36 locations, 8 years and 3 target populations of environments (TPE) in India. Four prediction scenarios and 6 kernel-models within/across TPEs were tested. Our results suggest that the conventional GBLUP, without enviromic data or when omitting EPA, is inefficient in predicting the performance of wheat lines in future years. However, when EPA was introduced as an intermediary learning step to reduce the dimensionality of the GxE kernels while connecting phenotypic and environmental-wide variation, a significant enhancement of GxE prediction accuracy was evident. EPA revealed that the effect of seasonality makes strategies such as 'covariable selection' unfeasible because GxE is year-germplasm specific. We propose that the EPA effectively serves as a 'reinforcement learner' algorithm capable of uncovering the effect of seasonality over the reaction norms, with the benefits of better forecasting the similarities between past and future trialing sites. EPA combines the benefits of dimensionality reduction while reducing the uncertainty of genotype-by-year predictions and increasing the resolution of GP for the genotype-specific level.

Prediction of near‐term climate change impacts on UK wheat quality and the potential for adaptation through plant breeding

“…Preliminary analysis investigated performing feature selection using gradient boosting machine learning models (Friedman, 2002 ) to select the most relevant ECs for each trait but this was not found to increase within‐trial prediction accuracy. This was supported by Costa‐Neto et al ( 2022 ) who also advocated using all ECs in ERM for prediction of new genotypes in new years. Therefore, a non‐linear kernel was calculated using all ECs according to Costa‐Neto, Fritsche‐Neto, et al ( 2021 ) and Costa‐Neto, Galli, et al ( 2021 ): where is the scaled EC matrix, is the Euclidean Distance between each element of the EC matrix ( n environments × m ECs), is a scaling factor assumed as the mean value of the Euclidean distance matrix and is a bandwidth factor which was assumed to be 1 as default.…”

“…This was supported by Costa-Neto et al (2022) who also advocated using all ECs in ERM for prediction of new genotypes in new where w is the scaled EC matrix, ‖w − w � ‖ 2 is the Euclidean Distance between each element of the EC matrix (n environments × m ECs), is a scaling factor assumed as the mean value of the Euclidean distance matrix and h is a bandwidth factor which was assumed to be 1 as default. The ERM therefore had dimensions of n environments × n environments.…”