An empirical Bayes method for robust variance estimation in detecting DEGs using microarray data

The microarray technology is widely used to identify the differentially expressed genes due to its high throughput capability. The number of replicated microarray chips in each group is usually not abundant. It is an effcient way to borrow information across different genes to improve the parameter estimation which suffers from the limited sample size. In this paper, we use a hierarchical model to describe the dispersion of gene expression profiles and model the variance through the gene expression level via a link function. A heuristic algorithm is proposed to estimate the hyper-parameters and link function. The differentially expressed genes are identified using a multiple testing procedure. Compared to SAM and LIMMA, our proposed method shows a significant superiority in term of detection power as the false discovery rate being controlled.