Dear Michael I second Ryan's advice. Rationale: unless there is a catastrophic data quality problem, then the variability from the technical replicates only reflects the (more or less Poissonian) counting noise, and is appropriately taken care of by just adding the counts. The counting noise is small compared to the biological variability for all but the genes with the lowest counts, say below ~200. OTOH, if there is a catastrophic data quality problem, then it's better to just drop the affected sample / lane / library. Best wishes Wolfgang El May 1, 2013, a las 8:25 pm, Ryan Thompson <rct at="" thompsonclan.org=""> escribi?: > I believe that the simplest way to deal with technical replicates is to > simply add their counts together, so that you have one column for each > biological replicate. > > On Wednesday, May 1, 2013, Michael Muratet wrote: > >> Hi Simon >> >> I've been using DESeq2 to analyze RNA-seq data I was given for a >> multi-factor experiment, three factors with two, three and three levels >> each with three 'replicates' in each cell. I recently learned that the >> replicates are actually technical, not biological, and I'm looking for the >> best way to set up the design matrix to take the correlation into account. >> >> I don't see anything in the manual or on the list, I'd appreciate your >> input. Should I use a weighting scheme via normalization factors? >> >> Thanks >> >> Mike >> >> >> Michael Muratet, Ph.D. >> Senior Scientist >> HudsonAlpha Institute for Biotechnology >> mmuratet at hudsonalpha.org <javascript:;> >> (256) 327-0473 (p) >> (256) 327-0966 (f) >> >> Room 4005 >> 601 Genome Way >> Huntsville, Alabama 35806 >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor at r-project.org <javascript:;> >> https://stat.ethz.ch/mailman/listinfo/bioconductor >> Search the archives: >> http://news.gmane.org/gmane.science.biology.informatics.conductor >> > > [[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor at r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor

On 01/05/13 20:25, Ryan Thompson wrote: > I believe that the simplest way to deal with technical replicates is to > simply add their counts together, so that you have one column for each > biological replicate. > > On Wednesday, May 1, 2013, Michael Muratet wrote: > I've been using DESeq2 to analyze RNA-seq data I was given for a > multi-factor experiment, three factors with two, three and three > levels each with three 'replicates' in each cell. I recently learned > that the replicates are actually technical, not biological, and I'm > looking for the best way to set up the design matrix to take the > correlation into account. Ryan is correct: adding up the technical replicates is the way to go. As your design is two-way, you will (hopefully) still have enough degrees of freedom left to estimate the dispersion. Simon