Hello all, I have some questions regarding whether the significant dye bias I'm finding in my analyses could be an artefact of my analysis method. I've been using limma to analyse a simple design comparing treatment and control cases using dye swaps. As per suggestions in the recent limma Users' Guide, I've added an intercept term to the design, and used it to find genes with significant dye effects. limma reports very many significantly dye- biased genes (B-values as high as 12.7, 205 genes with B > 5), and very few significantly differentially-expressed genes (highest B = 3.1). I'm using three biological replicates, each hybridised to two dye- swapped arrays as technical replicates, on Compugen human 19k cDNA slides. Is such a strong result plausible, or due to me incorrectly analysing the data? If so, what major pitfalls could I have blundered into? What sort of diagnostics can I try to test how reliable the model results are? Thanks for your time, Mark Pinese

Hi Mark. A way of controling your model would be to plot the genes that you find to have high b-scores due to dye bias in RI-plots (MA-plots). If they always tend to have, say a positive M-value, in one of the dyes (regardless of control of treatment), say the green dye, then you probably have some genes differentially expressed due to dye bias. Best regards // Johan -----Original Message----- From: bioconductor-bounces@stat.math.ethz.ch [mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of Mark Pinese Sent: Wednesday, July 20, 2005 12:21 AM To: bioconductor at stat.math.ethz.ch Subject: [BioC] Significant dye bias using limma Hello all, I have some questions regarding whether the significant dye bias I'm finding in my analyses could be an artefact of my analysis method. I've been using limma to analyse a simple design comparing treatment and control cases using dye swaps. As per suggestions in the recent limma Users' Guide, I've added an intercept term to the design, and used it to find genes with significant dye effects. limma reports very many significantly dye-biased genes (B-values as high as 12.7, 205 genes with B > 5), and very few significantly differentially-expressed genes (highest B = 3.1). I'm using three biological replicates, each hybridised to two dye-swapped arrays as technical replicates, on Compugen human 19k cDNA slides. Is such a strong result plausible, or due to me incorrectly analysing the data? If so, what major pitfalls could I have blundered into? What sort of diagnostics can I try to test how reliable the model results are? Thanks for your time, Mark Pinese _______________________________________________ Bioconductor mailing list Bioconductor at stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/bioconductor

I guess the idea is that, as you have included dye-bias in your model, you can now judge the effects of treatment with impunity. If you hadn't included it in your model, then any "treatment" effects you observed/were reported *could* have been due to dye-effects. BUT then, you wouldn't have known your array had significant dye-effects, and therefore you wouldn't have cared :-p Have you looked at the original data? If you have technical (or biological) replicates as dye-swaps, what do the numbers look like? Is there a good correlation? -----Original Message----- From: bioconductor-bounces@stat.math.ethz.ch [mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of Mark Pinese Sent: 21 July 2005 02:08 To: bioconductor at stat.math.ethz.ch Subject: Re: [BioC] Significant dye bias using limma Will such a significant bias affect the validity of my treatment effect results? In other words, can I just appreciate that dye bias is rampant, then ignore it and confidently extract meaningful statistics from my treatment vs control coefficient? Mark Gordon K Smyth wrote: >The fact that the dye effect is often highly significant is the reason >that it is recommended to include it in the model. > >Gordon > > >> >>Is such a strong result plausible, or due to me incorrectly analysing >>the data? If so, what major pitfalls could I have blundered into? >>What sort of diagnostics can I try to test how reliable the model >>results are? >> >> >> _______________________________________________ Bioconductor mailing list Bioconductor at stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/bioconductor