Genome-wide association study for non-normally distributed traits: A case study for stalk lodging in maize

Genome-wide association study;Maize;Plant breeding;Quantitative genetics

Shenstone, Esperanza M. ; Lipka ; Alexander E.

关键词: Genome-wide association study;    Maize;    Plant breeding;    Quantitative genetics;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/99387/SHENSTONE-THESIS-2017.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

The abundance of new genomic information available has increased the ability of computational tools to study the genetic basis of agricultural traits, notably with the application of the Genome-Wide Association Study (GWAS). A limitation of GWAS is that the assumptions underlying the linear model typically used to conduct the analysis are often violated in nature, and in such cases, the linear model is inappropriate to use. Alternatively, the mixed logistic regression model is well-suited for a genome-wide association study of binomially distributed agronomic traits because it can include fixed and random effects that account for spurious associations. However, the computational burden associated with fitting this model renders it inefficient to use at every genetic marker that are analyzed in the genome-wide association study.Therefore, the purpose of this work was to assess the ability of simpler statistical models to identify promising subsets of genome-wide markers to apply to the mixed logistic regression model. We tested this approach on stalk lodging, a binomially distributed trait measured on a maize (Zea mays L.) diversity panel. This analysis culminated in the mixed logistic regression model identifying genomic regions coinciding with signals associated with closely related quantitative traits. Using genomic data from the same panel, we conducted a simulation study to determine which parameters of the binomial distribution most likely contribute to the detection of quantitative trait nucleotides. The results suggest that the discovery of such signals is maximized when the probability of a successful Bernoulli trial is 0.5. Based on our findings, we present an analytical framework that involves phenotyping binomially distributed traits so that the possibility of identifying associated markers is maximized and then prioritizes subsets of genome-wide markers for fitting the mixed logistic regression model; such prioritization should make it practical to use the mixed logistic regression model to test for marker-trait associations on an average computer.