Hello Users, I have a question regarding the usage of backgroundCorrect function in LIMMA. when I do the following with offset 50, I am getting 2900 differentially expressed genes RG.b <- backgroundCorrect(RG, method = "normexp", offset = 50); where as, when I do the following with offset 1, I am getting 1300 differentially expressed genes RG.b <- backgroundCorrect(RG, method = "normexp", offset = 1); Please advise which offset value to be used? Why is offset value making so much difference? I am using this for TWO channel data, which is read by "genepix". Greatly appreciate your help. Prasad

Hi Prasad, On 8/26/2011 11:00 AM, Prasad Siddavatam wrote: > > > Hello Users, > > I have a question regarding the usage of backgroundCorrect function in LIMMA. > > when I do the following with offset 50, I am getting 2900 differentially > expressed genes > RG.b<- backgroundCorrect(RG, method = "normexp", offset = 50); > > where as, when I do the following with offset 1, > I am getting 1300 differentially expressed genes > RG.b<- backgroundCorrect(RG, method = "normexp", offset = 1); > > Please advise which offset value to be used? Why is offset value making > so much difference? I can't advise you on the offset to use; that is up to you as the data analyst. But I can explain why you get more genes with a larger offset. When you do a local background correction of your data, for the set of spots that are fairly dim (not much different from background intensity), the resulting ratios can become unstable because the numerators and/or denominators get small. This gives the characteristic spreading of the MA plot at low intensities after background correction. An extreme example would be the instance where the R and G channels are nearly identical (say, 200 and 205), so the uncorrected ratio is close to 1. But if the Rb and Gb values are, say 190 and 185, then the background corrected ratio will be 2! Adding 50 to the R and G values before background correction will dampen the ratio to 0.86, which is likely closer to truth. If you do MA plots before background correction and then after, both with and without adding the offset you will see what I mean. Best, Jim > > I am using this for TWO channel data, which is read by "genepix". > > Greatly appreciate your help. > > Prasad > > _______________________________________________ > 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 -- James W. MacDonald, M.S. Biostatistician Douglas Lab University of Michigan Department of Human Genetics 5912 Buhl 1241 E. Catherine St. Ann Arbor MI 48109-5618 734-615-7826 ********************************************************** Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues

Hi Jim, Thank you very much for you explanation. Now I can understand the reason. But when you said its up to the analyst to decide on the offset value. Is this statement based on the number of genes I am expecting (to be differentially expressed) or on some other criteria. Its is very critical because I am using several types of arrays (agilent, cDNAs) Appreciate your help. Prasad

Hi Jim, I really really appreciate your help. But I have a problem here. The datasets are downloaded from NCBI. I can't get many details about the experiments, hence the trouble. Even the original publications doesn't say much about the experiments. Prasad

Dear Dr. Smyth, Thank you very much for your detailed response. I am going to read the references. I understand why I am getting more differential genes with bigger offset value. Basically we are bringing the variance close to zero (with more uniformity across probes). Are you suggesting to delete the probes before doing the background correct? If yes, is there any limit on the maximum number of probes to be deleted? In the variance stabilization, which maximum value (approximately) of fit$df.prior is considered high enough to call a good variance stabilization is achieved. Appreciate your time and help. Prasad

Dear Wei, Thank you very much. I will read the reference. Prasad

Dear Prasad, > Date: Mon, 29 Aug 2011 02:17:01 +0000 > From: Prasad Siddavatam <siddavatam at="" gmail.com=""> > To: <bioconductor at="" stat.math.ethz.ch=""> > Subject: Re: [BioC] backgroundCorrect offset value > > Dear Dr. Smyth, > > Thank you very much for your detailed response. I am going to read the > references. I understand why I am getting more differential genes with bigger > offset value. Basically we are bringing the variance close to zero (with more > uniformity across probes). > > Are you suggesting to delete the probes before doing the background correct? No, I'm not. All probes should be retained for background correction. Non-expressed probes should be filtered before using eBayes(). > If yes, is there any limit on the maximum number of probes to be deleted? > > In the variance stabilization, which maximum value (approximately) of > fit$df.prior is considered high enough to call a good variance stabilization is > achieved. There is no maximum value. Higher is better. I think you're worrying about this more than is necessary. A decent value like offset=50 will give good results in a wide variety of situations. You can even use fit <- eBayes(fit, trend=TRUE) which will makes uniformity of the variance less important. Again, use plotSA(fit) to see what this does. Best wishes Gordon > Appreciate your time and help. > > Prasad ______________________________________________________________________ The information in this email is confidential and intend...{{dropped:4}}