|
Abstract:
|
New age genetics aim to reveal underlying mechanisms that link genotypic variations with multiple phenotypes. This requires new computational tools to model high dimensional datasets generated by High-throughput screening. We propose an efficient framework to model the correlations among multiple phenotypes as a function of candidate genes and provide estimation methodology and inference of model parameters. Previously introduced methods on covariance regression (Pourahmadi, 1999; Hoff and Niu, 2012) are computationally intensive, and the results are difficult to interpret within the biological paradigm. To overcome these limitations, we modeled the standard deviation and pairwise correlation directly by maximizing the pairwise composite likelihood. We deployed an efficient estimation technique aided by the Minorize-Maximize (MM) algorithm. When compared to the existing estimation techniques, our method reduces the computational time by several folds and is thus particularly suitable for large scale data. We identified novel phenotypic patterns through multiple synthetic datasets and biologically verified the findings by analyzing a population of recombinant inbred lines of cowpea.
|
