A reaction norm model for genomic selection using high-dimensional genomic and environmental data.

Jarquín D, Crossa J, Lacaze X, Du Cheyron P, Daucourt J, Lorgeou J, Piraux F, Guerreiro L, Pérez P, Calus M, Burgueño J, de los Campos G. · 2014

Read A reaction norm model for genomic selection using high-dimensional genomic and environmental data. by Jarquín D, Crossa J, Lacaze X, Du Cheyron P, Daucourt J, Lorgeou J, Piraux F, Guerreiro L, Pérez P, Calus M, Burgueño J, de los Campos G. (2014) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3931944.txt.

Introduction

Diego Jarquín and colleagues present a statistical model for picking better crop lines when genes and weather interact, in Theoretical and Applied Genetics in 2014. A wheat line's yield depends on its genes and on the fields where it grows, and the best genes in one place may not win in another. Genotyping now yields thousands of markers, and weather records are just as dense, so modeling every gene-by-weather pairing directly is impossible.

Their reaction norm model handles the flood by working with similarity patterns among lines and among environments instead of single pairings. Tested on wheat, it predicted grain yield more accurately, both across different environments and within one environment, than models without the interaction terms. Breeders gain a workable way to use genomic and weather data together.

Catalog record

CREATORJarquín D, Crossa J, Lacaze X, Du Cheyron P, Daucourt J, Lorgeou J, Piraux F, Guerreiro L, Pérez P, Calus M, Burgueño J, de los Campos G.
YEAR2014
COLLECTIONPAPERS
FORMATTXT
FILE SIZE91 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC3931944

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A reaction norm model for genomic selection using high-dimensional genomic and environmental data By Diego Jarquín, José Crossa, Xavier Lacaze, Philippe Du Cheyron, Joëlle Daucourt, Josiane Lorgeou, François Piraux, Laurent Guerreiro, Paulino Pérez, Mario Calus, Juan Burgueño, Gustavo de los Campos Published in: TAG. Theoretical and Applied Genetics. Theoretische Und Angewandte Genetik DOI: 10.1007/s00122-013-2243-1 ABSTRACT Key message New methods that incorporate the main and interaction effects of high-dimensional markers and of high-dimensional environmental covariates gave increased prediction accuracy of grain yield in wheat across and within environments. Abstract In most agricultural crops the effects of genes on traits are modulated by environmental conditions, leading to genetic by environmental interaction (G × E). Modern genotyping technologies allow characterizing genomes in great detail and modern information systems can generate large volumes of environmental data. In principle, G × E can be accounted for using interactions between markers and environmental covariates (ECs). However, when genotypic and environmental information is high dimensional, modeling all possible interactions explicitly becomes infeasible. In this article we show how to model interactions between high-dimensional sets of markers and ECs using covariance functions. The model presented here consists of (random) reaction norm where the genetic and environmental gradients are described as linear functions of markers and of ECs, respectively. We assessed the proposed method using data from Arvalis, consisting of 139 wheat lines genotyped with 2,395 SNPs and evaluated for grain yield over 8 years and various locations within northern France. A total of 68 ECs, defined based on five phases of the phenology of the crop, were used in the analysis. Interaction terms accounted for a sizable proportion (16 %) of the within-environment yield variance, and the prediction accuracy of models including interaction terms was substantially higher (17–34 %) than that of models based on main effects only. Breeding for target environmental conditions has become a central priority of most breeding programs. Methods, like the one presented here, that can capitalize upon the wealth of genomic and environmental information available, will become increasingly important. INTRODUCTION In the analysis of agricultural data and plant breeding experiments, the development of methods for modeling the interaction between genotypes and environments (G × E) precedes the development of analysis of variance. For instance, Fisher and Mackenzie (1923) suggested modeling the differential responses of genotypes (G) to environments (E) using a multiplicative (product) operator rather than additive models. Yates and Cochran (1938) proposed using a multiplicative operator consisting of a simple regression of a line’s performance on the environmental mean (joint-regression analysis). Years later, other multiplicative operators based on singular value decomposition were proposed and used by Gollob (1968), Mandel (1969), Gauch (1988), Cornelius et al. (1996) and Crossa and Cornelius (1997). Later, Piepho (1998) and Smith et al. (2001, 2005) used this multiplicative operator for modeling G × E but in the context of linear mixed-effect models, and Crossa et al. (2004, 2006) and Burgueño et al. (2008, 2011) considered the use of structured covariance matrices to model G × E in the context of pedigree-based mixed models. The main effects of genes and of environmental conditions could be modeled by regressing phenotypes on genetic markers and on environmental covariates (ECs; e.g., temperature, soil moisture, solar radiation) concurrently; and G × E can in principle be modeled using interactions between genetic markers and ECs. An example of such an approach is the factorial regression (FR) model (Denis 1988; van Eeuwijk et al. 1996; Vargas et al. 1999, 2001). Models for QTL × environment interaction (Q × E) have been applied both in the context of fixed-effects regression, such as the FR, and using Partial Least Squares (Crossa et al. 1999; Vargas et al. 2006). More recently, these methods were used in a mixed model context (Boer et al. 2007; Malosetti et al. 2004) and later extended to multi-environment multi-trait model settings (Malosetti et al. 2008; Alimi et al. 2013). With the development of modern genotyping and sequencing technologies, molecular marker information has become high dimensional, with the number of markers (p) potentially exceeding by large the number of phenotypic records (n) available for model fitting. Similarly, as climatic and agronomic information systems develop, environmental information also becomes high dimensional. The Q × E models discussed above cannot cope with the high-dimensional nature of genomic and EC information. A simple approach would be to introduce a first step where a few ‘significant’ ECs and genomic regions are selected, with only a small subset of markers and ECs used in the final model. However, with variable selection procedures important volumes of information can be lost in the process of selecting markers (Bernardo 2010; Crossa 2012) or ECs, with the undesirable result that large proportions of genetic or environmental signals may be unaccounted for. To circumvent the limitations of QTL-based models, Meuwissen et al. (2001) proposed using whole genome regression (WGR) methods (also known as models for genomic selection, GS) where information about potentially hundreds of thousands of markers is jointly considered. Such models allow capturing not only major-effect genes but also the contribution of genomic regions with small effects. Implementing these large-p with small-n regressions is possible using modern shrinkage estimation procedures, and empirical evidence obtained with plant breeding data shows that GS can outperform the predictive power of pedigree-based methods by a sizable amount (de los Campos et al. 2009; Crossa et al. 2010, 2011; Heslot et al. 2012; Pérez et al. 2010). Recently, GS models were extended to multi-trait multi-environment settings. For instance, Burgueño et al. (2012) used a multi-environment version of the genomic best linear unbiased predictor (G-BLUP) where G × E was modeled using genetic correlations, and found that the multi-environment G-BLUP had a much higher prediction accuracy than the single-trait G-BLUP. However, the study of Burgueño et al. (2012) did not incorporate environmental variables to model G × E. The principles used in GS to model the effects of genetic markers could also be exploited for modeling the main effects of large numbers of ECs, and, in theory, one could also include in the model explicitly all possible contrasts between markers and ECs. However, such an approach would be extremely demanding because the number of contrasts to be considered (and consequently the number of effects to be estimated) grows proportionally to the product of the number of markers and the number of ECs, leading to important statistical and computational challenges. In this article, we propose a class of random effects models where the main and interaction effects of markers and ECs are introduced using covariance structures that are functions of marker genotypes and ECs. The proposed approach represents an extension of the G-BLUP and can be interpreted as reaction norm model (e.g., Woltereck 1909; Gregorius and Namkoong 1986; Falconer and Mackay 1996; Calus et al. 2002; Calus and Veerkamp 2003; Su et al. 2006) where genetic and environmental gradients are described using a linear regression on genetic markers and on ECs. We evaluated the proposed methods using a data set from Arvalis, consisting of 139 wheat lines evaluated for grain yield in 340 year × location combinations. Genetic and ECs information consisted of 2,395 SNPs and 68 ECs. The prediction accuracy of the proposed models was assessed for two

Read the complete text