ModellingG×Ewith historical weather information improves genomic prediction in new environments

Abstract

Interaction between the genotype and the environment (G×E) has a strong impact on the 7 yield of major crop plants. Although influential, taking G×E explictily into account in plant 8 breeding has remained difficult. Recently G×E has been predicted from environmental and 9 genomic covariates, but existing works have not shown that generalization to new environ-10 ments and years without access to in-season data is possible and practical applicability re-11 mains unclear. Using data from a Barley breeding program in Finland, we construct an 12 in-silico experiment to study the viability of G×E prediction under practical constraints. We 13show that the response to the environment of a new generation of untested Barley cultivars 14 can be predicted in new locations and years using genomic data, machine learning and his-15 torical weather observations for the new locations. Our results highlight the need for models 16 of G×E: non-linear effects clearly dominate linear ones and the interaction between the soil 17 type and daily rain is identified as the main driver for G×E for Barley in Finland. Our study 18 implies that genomic selection can be used to capture the yield potential in G×E effects for 19 future growth seasons, providing a possible means to achieve yield improvements, needed for 20 1 . CC-BY-NC-ND 4.0 International license It is made available under a was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.The copyright holder for this preprint (which . http://dx.doi.org/10.1101/213231 doi: bioRxiv preprint first posted online Nov. 2, 2017; feeding the growing population. 21Global yield improvements are needed to feed the growing population the genotype and environment could be realised. While the importance of G×E for agronomic 25 performance is widely accepted, utilisation calls for methods that predict yields in new environ-26 ments, because actual experimental data, consisting of yields of plant variety candidates from yield 27 trials, will in practice be available from only a very limited number of environments. Importantly, 28prediction of a plant's response to a new environment cannot be based on weather data from the 29 growth season, as those will never be available at the time of prediction. 2. CC-BY-NC-ND 4.0 International license It is made available under a was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.The copyright holder for this preprint (which . http://dx.doi.org/10.1101/213231 doi: bioRxiv preprint first posted online Nov. 2, 2017; Figure 1: Outline of our approach. a) Precision breeding aims at producing varieties that are optimal for a specific environment. As compared to traditional breeding (b), targeted breeding aims at higher environmental adaptation, i.e., smaller target environments. Weather (microclimate) is a crucial driver for agronomic performance, but as it is unknown for future growth seasons, we use historical weather records (c) ...

Enviromic Assembly Increases Accuracy and Reduces Costs of the Genomic Prediction for Yield Plasticity in Maize

“…Conversely, environmental typologies (T) are based on frequencies (ranging from 0 to 1), where the sum of all frequencies is equal to 1 (100% of the variation). In addition, these typologies can be built for a given site using historical weather data, adapting the approach of Gillberg et al (2019) and de los Campos et al (2020). As presented in section GBLUP with enviromic main effects from T matrix (E-GP), if no typologies are considered, the expected environment effect is given for a fixed-environment intercept (with 0 variances within and between environments).…”

“…Consequently, it generates the well-reported lack of accuracy under genotype × environment interaction (G×E) conditions (Crossa et al, 2017). Therefore, novel ways that include environmental data (Heslot et al, 2014;Jarquín et al, 2014;Ly et al, 2018;Gillberg et al, 2019;Millet et al, 2019;Monteverde et al, 2019;Costa-Neto et al, 2021a) and process-based crop growth models (CGMs) (Messina et al, 2018;Robert et al, 2020;Toda et al, 2020; are considered the best pathways to fix it in the context of the multienvironmental GP. Most of the success of such approaches lies in understanding the ecophysiology interplay between genomics diversity and environment variation (Gage et al, 2017;Li et al, 2018;Guo et al, 2020;Costa-Neto et al, 2021b).…”

Using Local Convolutional Neural Networks for Genomic Prediction

“…Thus, enabling the use of genetic evaluations in large scale genetic evaluations such as the US Holstein population with 569,404 genotyped animals (Masuda et al, 2016). Similarly, computing time and memory requirements tend to explode in state-of-the-art methods such as GBLUP when additional input dimensions like weather data (Gillberg et al, 2019) are considered or multi-trait models are used.…”

Optimizing Genomic-Enabled Prediction in Small-Scale Maize Hybrid Breeding Programs: A Roadmap Review

“…Combined with phenotypic and genotypic data, the use of envirotypic data sources can leverage the molecular breeding strategies addressing the prediction of tested and untested environments, such as climate change scenarios (Millet et al, 2016(Millet et al, , 2019Messina et al, 2018;Bustos-Korts et al, 2019;de los Campos et al, 2020;Guo et al, 2020). These data have been incorporated into GP in the last ten years to better model the G × E interaction according to the reaction norm (Heslot et al, 2014;Jarquín et al, 2014;Gillberg et al, 2019;Costa-Neto et al, 2020;Rogers et al, 2021). However, it is difficult for most breeders to deal with this interaction between environmental models, ecophysiology, and genetics (Costa-Neto et al, 2021), in which we need to (i) implement a cost-effective and intuitive pipeline to integrate envirotyping data in GP and (ii) find novel enviromic approaches, more capable of describing phenotype-envirotype covariances and translate it into accuracy gains.…”

“…Finally, we find that using multi-trait multi-environment data might help design better field phenotyping trials for training GP models. As the modern computational tools attempt better to explore G × E and G × G within a multi-environment multi-trait context, the opposite path might be taken by using historical data to design future trials (Rincent et al, 2017) and scenarios (Millet et al, 2016;Bustos-Korts et al, 2019), but also to predict cultivars at novel growing conditions (Gillberg et al, 2019;Millet et al, 2019;de los Campos et al, 2020).…”