To those who responded to my last e-mail, thanks for the help. I had another question. I got my 2 color time course data into limma. I have a targets file with 2 replicates per time point for time points 1 day, 2 day, 4 day, 7 day, and 14 day. I have LIMMA assuming that these are not ALL replicates by telling it so. Please not that any semicolons coming up are NOT part of the code. There is supposed to be a nicely sized shift in differential expression from 4 days to 7 days, so I used those points for my comparison. As the LIMMA manual has stated, I have assigned my levels variable lev, assigned my factors variable f, and my design. I made my colnames variable: colnames(design)=lev ; and my fit variable: fit=lmFit(MA, design); where MA is the normalized RG. I continue to follow the manual (the variable names it gave me were X1day, X2day, etc.): cont=makeContrasts("X7day-X4day", levels=design); I then did fit2=contrasts.fit(fit, cont) ; then fit2=eBayes(fit2); then I did selected=p.adjust(fit2$F.p.value, method="BH")<0.05 to get the genes that change from 4 days to 7 days with strong p-values. Unfortunately, looking at the results yield only about 30 genes (there should be several hundred), none of whom (by eye) undergo any significant change in differential expression from the 4 day point to the 7 day point. Can someone please help me with what I may be doing wrong? Any help would be greatly appreciated. Thanks! -greg

Just to add a bit here, with many image analysis options, there are other measures of the "quality" of a spot besides intensity. Since limma will allow you to incorporate this information into an analysis, you might think about whether there is some quantity reported by your image analysis software that might be useful in this regard. I agree with Naomi that excluding genes based on low-intensity spots in some subset of the arrays may be discarding some of the most interesting data. Sean On 7/19/06 8:10, "Naomi Altman" <naomi at="" stat.psu.edu=""> wrote: > I would not do this. Use less background correction, (e.g. don't > background correct, or subtract 1/2 of the background), or set the > channels that are below background to some low value (e.g. 1) so that > logs can be used. > > --Naomi > > At 09:48 AM 7/19/2006, you wrote: >> Thanks for your quick response. I will not delete the gene completely (if you >> delete genes then LIMMA doesn't know how to handle genes lists with different >> orders), but although it is helpful to keep genes that may have >> information in >> one array, I do think it may be necessary to "NA" the below background values >> and keep the above background ones. Thus you still have the good values but >> have eliminated possible bad ones. What do you think of this? >> -greg >> >> Quoting Naomi Altman <naomi at="" stat.psu.edu="">: >> >>> I would not delete data that is below background, even in both >>> channels, if it is above background on at least one array. >>> >>> It seems to me that it is important information to know that a gene >>> >>> does not express under some condition in your experiment. Of course, >>> >>> the unfortunate side-effect of our liking to use ratios is that >>> "zero" is not handled well. But a gene that expresses in some >>> conditions of interest but not in others surely is of primary >>> interest to your study. >>> >>> --Naomi >>> >>> At 11:48 AM 7/18/2006, milesg at bu.edu wrote: >>>> HI, my name is Gregory Miles. I'm at Boston University and was given >>> this >>>> address by Dr. Carey (I went to a seminar of his last week) at the >>> Harvard >>>> medical school and was told that I could ask my question about 2 >>>> color data to >>>> you. On the mouse microarray dataset we have, there are two colors, >>> and >>>> therefore two values that can be below background. When both values >>> are above >>>> background (zero_barcode on our chip), we keep the data and when >>> both are >>>> below we eliminate the data (they become NA). I imagine this is a >>> correct >>>> approach, but what should be done regarding the data that has one >>> intensity >>>> below background and one above. Would it be best to keep the good >>>> value? Do we >>>> eliminate the entire gene from entry into bioconductor? Perhaps >>>> there is a way >>>> to specify to bioconductor that this is the case (by entering a >>> background >>>> value) and allowing it to handle the data abstractly? Or is it best >>> to let >>>> Bioconductor look at them as NA's. Any help would be greatly >>> appreciated. >>>> Thanks! >>>> -Greg Miles >>>> >>>> _______________________________________________ >>>> Bioconductor mailing list >>>> Bioconductor at stat.math.ethz.ch >>>> https://stat.ethz.ch/mailman/listinfo/bioconductor >>>> Search the archives: >>>> http://news.gmane.org/gmane.science.biology.informatics.conductor >>> >>> Naomi S. Altman 814-865-3791 (voice) >>> Associate Professor >>> Dept. of Statistics 814-863-7114 (fax) >>> Penn State University 814-865-1348 >>> (Statistics) >>> University Park, PA 16802-2111 >>> >>> > > Naomi S. Altman 814-865-3791 (voice) > Associate Professor > Dept. of Statistics 814-863-7114 (fax) > Penn State University 814-865-1348 (Statistics) > University Park, PA 16802-2111 > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor

Hi Miles, We are all trying to say: Why do you think a channel with a near- or below-background is a "bad" value and needs to be removed? If the gene is not expressed in one treatment, then it should have a 'zero' value, which due to array technology will be some positive number near background fluorescence. By removing the value completely, you are saying in the analysis "there is no information for that sample on that array", but you do have information on that sample - it was below detectable level. Even though the number that comes out of the background correction, either 0.5 or 1 as suggested, is not entirely accurate, it is relatively accurate to numbers a good deal higher. Conversely, you should not throw away saturated values either, because even though you don't know exactly how large they were, you do know they were large. If both channels of a spot are near/below background on every single array in your experiment, then you can remove the entire gene/spot from the analysis. Cheers, Jenny At 09:10 AM 7/19/2006, Naomi Altman wrote: >I would not do this. Use less background correction, (e.g. don't >background correct, or subtract 1/2 of the background), or set the >channels that are below background to some low value (e.g. 1) so that >logs can be used. > >--Naomi > >At 09:48 AM 7/19/2006, you wrote: > >Thanks for your quick response. I will not delete the gene completely > (if you > >delete genes then LIMMA doesn't know how to handle genes lists with > different > >orders), but although it is helpful to keep genes that may have > >information in > >one array, I do think it may be necessary to "NA" the below background > values > >and keep the above background ones. Thus you still have the good values but > >have eliminated possible bad ones. What do you think of this? > >-greg > > > >Quoting Naomi Altman <naomi at="" stat.psu.edu="">: > > > > > I would not delete data that is below background, even in both > > > channels, if it is above background on at least one array. > > > > > > It seems to me that it is important information to know that a gene > > > > > > does not express under some condition in your experiment. Of course, > > > > > > the unfortunate side-effect of our liking to use ratios is that > > > "zero" is not handled well. But a gene that expresses in some > > > conditions of interest but not in others surely is of primary > > > interest to your study. > > > > > > --Naomi > > > > > > At 11:48 AM 7/18/2006, milesg at bu.edu wrote: > > > >HI, my name is Gregory Miles. I'm at Boston University and was given > > > this > > > >address by Dr. Carey (I went to a seminar of his last week) at the > > > Harvard > > > >medical school and was told that I could ask my question about 2 > > > >color data to > > > >you. On the mouse microarray dataset we have, there are two colors, > > > and > > > >therefore two values that can be below background. When both values > > > are above > > > >background (zero_barcode on our chip), we keep the data and when > > > both are > > > >below we eliminate the data (they become NA). I imagine this is a > > > correct > > > >approach, but what should be done regarding the data that has one > > > intensity > > > >below background and one above. Would it be best to keep the good > > > >value? Do we > > > >eliminate the entire gene from entry into bioconductor? Perhaps > > > >there is a way > > > >to specify to bioconductor that this is the case (by entering a > > > background > > > >value) and allowing it to handle the data abstractly? Or is it best > > > to let > > > >Bioconductor look at them as NA's. Any help would be greatly > > > appreciated. > > > >Thanks! > > > >-Greg Miles > > > > > > > >_______________________________________________ > > > >Bioconductor mailing list > > > >Bioconductor at stat.math.ethz.ch > > > >https://stat.ethz.ch/mailman/listinfo/bioconductor > > > >Search the archives: > > > >http://news.gmane.org/gmane.science.biology.informatics.conductor > > > > > > Naomi S. Altman 814-865-3791 (voice) > > > Associate Professor > > > Dept. of Statistics 814-863-7114 (fax) > > > Penn State University 814-865-1348 > > > (Statistics) > > > University Park, PA 16802-2111 > > > > > > > >Naomi S. Altman 814-865-3791 (voice) >Associate Professor >Dept. of Statistics 814-863-7114 (fax) >Penn State University 814-865-1348 (Statistics) >University Park, PA 16802-2111 > >_______________________________________________ >Bioconductor mailing list >Bioconductor at stat.math.ethz.ch >https://stat.ethz.ch/mailman/listinfo/bioconductor >Search the archives: >http://news.gmane.org/gmane.science.biology.informatics.conductor Jenny Drnevich, Ph.D. Functional Genomics Bioinformatics Specialist W.M. Keck Center for Comparative and Functional Genomics Roy J. Carver Biotechnology Center University of Illinois, Urbana-Champaign 330 ERML 1201 W. Gregory Dr. Urbana, IL 61801 USA ph: 217-244-7355 fax: 217-265-5066 e-mail: drnevich at uiuc.edu