Session: Advanced Phenotypic Data Analyses: One- and Two-Stage Approaches and Genotype by Environment Interaction (GXE) Analysis

Division: Workshops

Title: Advanced Phenotypic Data Analyses: One- and Two-Stage Approaches and Genotype by Environment Interaction (GXE) Analysis

Lead Community Sponsor:
Cosponsor: ASA Section: Biometry and Statistical Computing, C01 Crop Breeding and Genetics
Community Cosponsor: Statistical Education/Training for Researchers Community, Bioinformatics in Crops and Soils Community
Format: Oral None (Admin Only)
Keywords:

Session Description: In the era of genomic selection and predictive analytics, the accurate analyses of phenotypic data will have a profound impact on the success of these technologies. Testing of breeding lines in multiple target environments is commonly done beginning with the preliminary and advanced trials in a cultivar development program. Additionally, testing of lines/germplasm at multiple locations and years is also necessary for trait characterization and for identifying the underlying genes. The trials conducted at multiple sites/years can vary in the experimental design, requirement of different spatial correction model at each site, heterogeneous variances at each site, and other factors. The multi-environment trials are analyzed using linear mixed models either in one-stage or two-stage analyses. For the one-stage analysis, unstructured and factor analytic models are commonly used for modeling genotype-by-environment (GxE) interaction patterns. However, due to the differences mentioned above and increased computing power needed for one-stage, a two-stage analysis may be necessary. A two-stage analysis will involve accounting for experimental design, covariates, and spatial variation at individual sites and getting the best phenotypic estimates at each location in the first-stage, and then combining these estimates across locations in the second-stage. Generally, to minimize the loss of information in a two-stage analysis, the inverse of the variances of predictions of values from the first-stage is used as weights in the second-stage. Further, the accurate phenotypic estimates derived from these analyses are critical for all the downstream applications where the phenotypic values are used. Although the benefits of these two approaches are now fairly-well known, little training is offered for these methods. Hence, this workshop will be for advanced users with a background in using linear mixed models. The workshop will be offered in an interactive format and will involve hands-on exercises.