Abstract
The effects of climate change create formidable challenges for breeders striving to produce sufficient food quantities in rapidly changing environments. It is therefore critical to investigate the ability of multi-environment genomic prediction (GP) models to predict genomic estimated breeding values (GEBVs) in extreme environments. Exploration of the impact of training set composition on the accuracy of such GEBVs is also essential. Accordingly, we examined the influence of the number of training environments and the use of environmental covariates (ECs) in GS models on four subsets of n = 500 lines of the soybean nested association mapping (SoyNAM) panel grown in nine environments in the US-North Central Region. The ensuing analyses provided insights into the influence of both of these factors for predicting grain yield in the most and the least extreme of these environments. We found that only a subset of the available environments was needed to obtain the highest observed prediction accuracies. The inclusion of ECs in the GP model did not substantially increase prediction accuracies relative to competing models, and instead more often resulted in negative prediction accuracies. Combined with the overall low prediction accuracies for grain yield in the most extreme environment, our findings highlight weaknesses in current GP approaches for prediction in extreme environments, and point to specific areas on which to focus future research efforts.
Highlights
The impacts of climate change are adversely affecting the availability of food, feed, fuel, and fiber security worldwide, with prior research suggesting a crop yield loss of 5% for each degree Celsius above historically observed weather patterns (Nelson et al, 2010; Zhao et al, 2017)
Because we were interested in the impact of training set composition on prediction accuracies in a given test environment, we evaluated the ability of all possible subsets of the remaining eight remaining environments to train each Genomic prediction (GP) model and accurately predict genomic estimated breeding values (GEBVs)
We evaluated the ability of M1–M3 to predict GEBVs in the most extreme environment, IA_2013, using the all-possible subsets of the 8 remaining environments, as described in the Section “Materials and Methods” and Table 1
Summary
The impacts of climate change are adversely affecting the availability of food, feed, fuel, and fiber security worldwide, with prior research suggesting a crop yield loss of 5% for each degree Celsius above historically observed weather patterns (Nelson et al, 2010; Zhao et al, 2017). Bernardo (1994) was the first who proposed the use of genomic information as covariates for predicting untested genotypes. Later on, Meuwissen et al (2001) proposed a new methodology to cope with the challenge of fitting prediction models when the number of genomic covariates (p), delivered with the advancements of sequencing technologies, surpass by far the number data points (n) available to fit models (p n)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have