Hi, I'm analysing data from a Affymetrix GeneChip Gene 1.0 ST microarray experiment. I'm analysing the data with the RMA method with the mixed model, nested ANOVA for the analysis of differential gene experiments. One of the experiments produces far fewer significant differential results than expected. I think this might be due to a lower than expected correlation between the replicates (three) of each control. To achieve results which might be closer to reality I was thinking of adjusting the FDR or p-value for calculation of differential gene expression. Currently I'm using 0.05 and 0.01 respectively. The Gene 1.0 ST chips contain probe sets selected by Affymetrix for their excellent performance, so I'm inclined to adjust the FDR. Would anybody have any further advice on this issue and/or papers which could inform my decision? Thanks
Can you provide more details please?
Are you profiling patient or cell line samples?
Have you carried out PCA or HCA?
Are you using gene or probeset normalisation?
How many gene are passing the threshold when you carry out straight differential expression as opposed to ANOVA?
What was the lowest FDR pvalue you are getting and the biggest fold change?
What software are you using?
I would be concerned if replicates do not cluster together. in addition either very small or big gene lists tend to indicate problems with the experiment.
you cannot rule out that there is was a problem in experimental design or a mistake in the lab.
while you can adjust the p-values to call differentially expressed genes - in practice wet lab follow up for potential targets the pvalue and fold changes are used to rank targets - there are other factor to consider i.e the underlying biology. of course if you want to use For example GSEA then you have to produce a defined list
I guess what I'm saying is you can change the cutoffs to produce a bigger list, but you increase the chance of hitting more false positives and that's not good for follow up analyses