|Title||Advantages of continuous genotype values over genotype classes for GWAS in higher polyploids : A comparative study in hexaploid chrysanthemum|
|Author(s)||Grandke, Fabian; Singh, Priyanka; Heuven, Henri C.M.; Haan, Jorn R. de; Metzler, Dirk|
|Source||BMC Genomics 17 (2016). - ISSN 1471-2164 - 9 p.|
|Department(s)||Animal Breeding and Genetics|
|Publication type||Refereed Article in a scientific journal|
|Keyword(s)||Association study - Bayz - Continuous genotypes - Linear regression - Partial least squares - Polyploids|
Background: Association studies are an essential part of modern plant breeding, but are limited for polyploid crops. The increased number of possible genotype classes complicates the differentiation between them. Available methods are limited with respect to the ploidy level or data producing technologies. While genotype classification is an established noise reduction step in diploids, it gains complexity with increasing ploidy levels. Eventually, the errors produced by misclassifications exceed the benefits of genotype classes. Alternatively, continuous genotype values can be used for association analysis in higher polyploids. We associated continuous genotypes to three different traits and compared the results to the output of the genotype caller SuperMASSA. Linear, Bayesian and partial least squares regression were applied, to determine if the use of continuous genotypes is limited to a specific method. A disease, a flowering and a growth trait with h 2 of 0.51, 0.78 and 0.91 were associated with a hexaploid chrysanthemum genotypes. The data set consisted of 55,825 probes and 228 samples. Results: We were able to detect associating probes using continuous genotypes for multiple traits, using different regression methods. The identified probe sets were overlapping, but not identical between the methods. Baysian regression was the most restrictive method, resulting in ten probes for one trait and none for the others. Linear and partial least squares regression led to numerous associating probes. Association based on genotype classes resulted in similar values, but missed several significant probes. A simulation study was used to successfully validate the number of associating markers. Conclusions: Association of various phenotypic traits with continuous genotypes is successful with both uni- and multivariate regression methods. Genotype calling does not improve the association and shows no advantages in this study. Instead, use of continuous genotypes simplifies the analysis, saves computational time and results more potential markers.