Abstract:
|
Heterosis, or hybrid vigor, is the enhancement of the phenotype of hybrid progeny relative to their inbred parents. To identify genes displaying a heterosis pattern in their expression, we construct a gene-specific overdispersed count regression model. Since there are ~40,000 genes and ~10 samples, we build a hierarchical count regression model that provides a data-based borrowing of information across genes. To implement a fully Bayesian analysis, we construct a novel parallelized Markov chain Monte Carlo algorithm that efficiently utilizes the architecture of a graphical processing unit through embarrassingly parallel computations and parallel reductions. We demonstrate the utility of the method to identify gene expression heterosis through a variety of simulation studies and analyze an RNA-seq maize dataset to identify genes with 6 different types of heterosis.
|