Hi Sean, Thank you very much for your reply. We have 12 arrays, 6 for each condition. If I understood correctly, you're suggesting to remove high variable genes within group (can i use 0.2% < CV < 10% to fiter out these high variable genes? ). I don't know how large for "large n"? Is it reasonable to use 6 per group? Can you suggest some literature on power calculation for microarray? I am having little experience with statistics. Thanks again, Regards, Sharon On 12/1/05, Sean Davis wrote: > > On 12/1/05 8:50 AM, "Sharon Anbu" wrote: > > > Hi All, > > > > I am working with the Affy data set for which the max fold change is > > around 1.7. Most of the probesets have a fold change between 1.2 and > > 1.5. I have done the standard data analysis procedure reccommanded by > > many people in Bioconductor (starting with RMA extraction, difference > > between 2-groups, t-test, p-val & corrected p-val etc). After doing > > this, I have got only 20 significant genes, which has no biological > > significance with the experiment. > > How many arrays have you run? If you know that you are looking for fold > changes of 1.2-1.7 (I'm not sure how you know which genes are true and > showing these changes, but...), you will likely need a stable system > (little > biological variability within groups) or a "relatively" large n. You > could > do a simple power calculation to determine what your power is to detect a > difference between your two groups given a "typical" variance and with > different numbers of slides. If you have good power with the number of > slides that you are using, then you may be seeing true negative > results. If > not, then increasing the number of arrays might be a good bet. All that > said, if you have a set of genes that you are interested in, you could go > directly from "fold-change" on the array (ignore statistics such as > p-value > altogether) and move directly to validation with PCR-based methods. > > Sean > > [[alternative HTML version deleted]]