A1175
Title: Meta-analytic study of experimental data in the presence of missingness and unbalanced experimental factors
Authors: Shyam Ranganathan - Clemson University (United States) [presenting]
Raghupathy Karthikeyan - Clemson University (United States)
Qiong Su - Clemson University (United States)
Abstract: Increased heat stress during cropping season poses significant challenges to rice production. A meta-analytic study of 1,946 experiments relating genetic factors (different rice species) was performed, and environmental factors (e.g., temperatures) to rice grain yield through random regression mixed models shows complex non-linear relationships. Rice yield is measured in terms of quantity components (panicle number, spikelet number per panicle, seed set rate, grain weight) and grain quality traits (milling yield, chalkiness, amylose, protein content). Model selection yields quadratic regression models,s and these models suggest both the optimum temperature ranges for high rice yields and the rice varieties most adaptive to temperature variations. However, there are significant variations across the large number of experiments used in the study due to missing data, variations in experimental conditions etc. While the number of experiments is large, each experiment includes few observations on a large number of variables, creating a high-dimensional problem. We propose a Bayesian framework to handle the meta-analysis in the presence of the limitations mentioned above. This will provide better results and help quantify the associated uncertainty. These results can be used to develop new field experiments to establish the relationships among phenotypic plasticity, genetic traits, and environmental factors.
