On 2013 Jun 28, Rafael Irizarry commented:

I highly recommend reading this paper. It presents the statistical motivation for the widely used limma package. Although originally developed for microarrays, I have used the ideas presented here for other high throughput data include next generation sequencing. Of particular importance is the idea of shrinking sample standard deviations before computing signal to noise summaries such as the t-test.

On 2013 Jun 28, Rafael Irizarry commented:

I highly recommend reading this paper. It presents the statistical motivation for the widely used limma package. Although originally developed for microarrays, I have used the ideas presented here for other high throughput data include next generation sequencing. Of particular importance is the idea of shrinking sample standard deviations before computing signal to noise summaries such as the t-test.