Greetings I am working on analyzing the data from an experiment with multiple technical replicates but no biological replicates. FWIW, it's the same data set I have posted about in recent weeks. I think the experiment would benefit greatly from some biological replicates. The question I have is this: given that the dispersion estimator algorithm is able to 'share' data among samples, does anyone know of any experiments or simulations to establish guidelines for a minimum or recommended number of biological replicates smaller than all N samples? The DESeq2 manual says that it employs Cox-Reid for estimating dispersions, is it same implementation as described in McCarthy et al., 2012, or are there some new twists? Thanks Mike Michael Muratet, Ph.D. Senior Scientist HudsonAlpha Institute for Biotechnology mmuratet at hudsonalpha.org (256) 327-0473 (p) (256) 327-0966 (f) Room 4005 601 Genome Way Huntsville, Alabama 35806

Hi Michael, On Jun 10, 2013 8:15 PM, "Michael Muratet" <mmuratet@hudsonalpha.org> wrote: > > Greetings > > I am working on analyzing the data from an experiment with multiple technical replicates but no biological replicates. FWIW, it's the same data set I have posted about in recent weeks. I think the experiment would benefit greatly from some biological replicates. The question I have is this: given that the dispersion estimator algorithm is able to 'share' data among samples, does anyone know of any experiments or simulations to establish guidelines for a minimum or recommended number of biological replicates smaller than all N samples? The power, or ability to detect true differential expression for a gene, is dependent on the sample size, but all also on the true log fold change. You could try using the makeExampleDESeqDataSet() function to simulate some data, which has a number of tunable parameters, in order to simulate different effect sizes and test power. >The DESeq2 manual says that it employs Cox-Reid for estimating dispersions, is it same implementation as described in McCarthy et al., 2012, or are there some new twists? The Cox Reid term is the same, but the method of information sharing across the gene estimates is different. This is described in the man page for estimateDispersions(). Mike [[alternative HTML version deleted]]