Hello, I've been running anova to identify modulated genes of an affy experiment (normalized data). When only considering those genes with a very confident F-test p-value (<=1e-4) there are hardly any genes found with very low intensities - which is good. My concern is/was that the anova analysis may pick up absent genes (genes not expressed at all - but for which there's a signal just due to noise). Surely this may happene if the anova p-value is raised. This is one reason why I'm a bit worried about using RMA since I cannot skip absent genes before I run the anova. Well, I can, but just based on the intensity, since RMA does not provide a p-value for the presence call liske MAS5. Does anova have the nice side effect to give bad (f-test) p-values to absent genes, because intesities just due to noise should have similar means and varaince in all factor levels? I'd be interested in your experience and opinion about this. kind regads, Arne -- Arne Muller, Ph.D. Toxicogenomics, Aventis Pharma arne dot muller domain=aventis com

We had a similar concern. We found that some genes known to express at very low levels in only one of our "treatments" in fact had a statistically significant limma F-value (which w consider to be a good thing). We did find some genes that appear to be statistically significant but which are expressed at a low level in all treatments. We do not have much information about how these genes should be expressed. Of course, there are always false positives and false negatives, and this cannot be avoided by statistical means. Limma or SAM appear to perform better than unadjusted ANOVA. --Naomi p.s. We did RT-PCR on several genes that were below our "detection threshold" but had significant MAS p-values and vice versa. Some of these genes were present at low levels and some were not. MAS did no better or worse than RMA at determining which genes were really absent. At 01:23 PM 6/24/2004 +0200, Arne.Muller@aventis.com wrote: >Hello, > >I've been running anova to identify modulated genes of an affy experiment >(normalized data). > >When only considering those genes with a very confident F-test p-value >(<=1e-4) there are hardly any genes found with very low intensities - >which is good. > >My concern is/was that the anova analysis may pick up absent genes (genes >not expressed at all - but for which there's a signal just due to noise). >Surely this may happene if the anova p-value is raised. > >This is one reason why I'm a bit worried about using RMA since I cannot >skip absent genes before I run the anova. Well, I can, but just based on >the intensity, since RMA does not provide a p-value for the presence call >liske MAS5. > >Does anova have the nice side effect to give bad (f-test) p-values to >absent genes, because intesities just due to noise should have similar >means and varaince in all factor levels? > >I'd be interested in your experience and opinion about this. > > kind regads, > > Arne > >-- >Arne Muller, Ph.D. >Toxicogenomics, Aventis Pharma >arne dot muller domain=aventis com > >_______________________________________________ >Bioconductor mailing list >Bioconductor@stat.math.ethz.ch >https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor Naomi S. Altman 814-865-3791 (voice) Associate Professor Bioinformatics Consulting Center Dept. of Statistics 814-863-7114 (fax) Penn State University 814-865-1348 (Statistics) University Park, PA 16802-2111