Bayesian-Variable-Selection

The genetic architecture of quantitative traits may benefit from greater shrinkage or variable selection on sequencing data depending on the distribution of true causal loci through a genome. Here I compare two prior distrubtions; the double expotential distribution which performs both shrinkage and variable selection on predictors, and the horseshoe prior mixture distrubtion with heavier variable selection. Additionally, I investigate the impact of secondary phenotypic traits derived from multispectral imagery to improve prediction accuracy with both priors.