Hello, I'm sorry if this question went to the wrong place, but I would be ecstatic if someone had the answer to it. I've been through most of the bioconductor site and online workshops and have read the vignettes and function help pages for all the relevant packages, but have been unable to find out how to use all the probe cells on affy chips in differential expression analysis packages in bioconductor. My data contains two groups with only two replicates each (and I am temporarily unable to come up with funds for the last two arrays), so it would be very beneficial for me to be able to use all 11 (or 16, etc.) probes in each array in order to get p-values, rather then just having two expression values to use in each group. Affymetrix's GCOS has the ability to perform Wilcoxon on these values for each probe set, but 1. I want to also perform quantile normalization on my data, and 2. I want to use empirical Bayes to determine the significance of differential expression calls, and GCOS can not do those. Also, GCOS can only compare two chips at a time. Here is what I know: The Limma package needs the data to be in an object of the class "exprset" with the samples as columns and the genes as rows. If I use the affy package and perform bg.correct() and normalization.AffyBatch.quantiles(), I can get my processed probe values. But now my probe values are lined up in columns, and I want to treat each probe in a probeset as a separate sample (so that I can get a p-value out of them), which means that I need the corresponding probes for each gene in one row, not 11 different rows. Since affy can find all the probes for a probeset (evidenced by both pmindex() and the fact that it can yield expression values), there should be a way to make matrices for all the arrays with the probe sets (instead of expression values) as rows that I could plug into limma. I'm aware that some probe sets have more probes than others, but would it work to fill in extra values as "NA" so that the matrix is "full"? And can anyone point me in the right direction as to how to create this matrix. Finally, where is the phenoData file that I would need to turn the matrix back into an exprset after it goes through limma? (answering the previous question will probably answer this as well). Sorry for the long windedness, and I would really be thankful for any help anyone could give. Thank You, David Young (from UNR)

Hi Regarding the complexity of probe-level analysis, I have a deposited article termed ResurfP in Genome Biology (http://genomebiology.com/2004/5/11/P14). I have not yet to made the software available to others (using Bioconductor tools or otherwise). If you would like to do such an analysis, you could contact me personally, for now. Suresh (Suresh Gopalan, Ph.D) ----- Original Message ----- From: "Gordon K Smyth" <smyth@wehi.edu.au> To: <davidl at="" unr.nevada.edu=""> Cc: <bioconductor at="" stat.math.ethz.ch=""> Sent: Monday, August 08, 2005 6:54 AM Subject: [BioC] Limma using all probes on Affy chip >> Date: Sun, 7 Aug 2005 20:37:22 -0700 >> From: davidl at unr.nevada.edu >> Subject: [BioC] Limma using all probes on Affy chip >> To: bioconductor at stat.math.ethz.ch >> >> Hello, >> >> I'm sorry if this question went to the wrong place, but I would be >> ecstatic >> if someone had the answer to it. I've been through most of the >> bioconductor >> site and online workshops and have read the vignettes and function help >> pages >> for all the relevant packages, but have been unable to find out how to >> use all >> the probe cells on affy chips in differential expression analysis >> packages in >> bioconductor. My data contains two groups with only two replicates each >> (and I >> am temporarily unable to come up with funds for the last two arrays), so >> it >> would be very beneficial for me to be able to use all 11 (or 16, etc.) >> probes >> in each array in order to get p-values, rather then just having two >> expression >> values to use in each group. Affymetrix's GCOS has the ability to >> perform >> Wilcoxon on these values for each probe set, > > This is not a valid analysis -- see below. > >> but 1. I want to also perform >> quantile normalization on my data, and 2. I want to use empirical Bayes >> to >> determine the significance of differential expression calls, and GCOS can >> not >> do those. Also, GCOS can only compare two chips at a time. Here is what >> I >> know: >> The Limma package needs the data to be in an object of the class >> "exprset" >> with the samples as columns and the genes as rows. If I use the affy >> package >> and perform bg.correct() and normalization.AffyBatch.quantiles(), I can >> get my >> processed probe values. But now my probe values are lined up in columns, >> and I >> want to treat each probe in a probeset as a separate sample (so that I >> can get >> a p-value out of them), which means that I need the corresponding probes >> for >> each gene in one row, not 11 different rows. Since affy can find all the >> probes for a probeset (evidenced by both pmindex() and the fact that it >> can >> yield expression values), there should be a way to make matrices for all >> the >> arrays with the probe sets (instead of expression values) as rows that I >> could >> plug into limma. I'm aware that some probe sets have more probes than >> others, >> but would it work to fill in extra values as "NA" so that the matrix is >> "full"? >> And can anyone point me in the right direction as to how to create this >> matrix. Finally, where is the phenoData file that I would need to turn >> the >> matrix back into an exprset after it goes through limma? (answering the >> previous question will probably answer this as well). Sorry for the long >> windedness, and I would really be thankful for any help anyone could >> give. >> >> Thank You, >> >> David Young (from UNR) > > The package you are looking is affyPLM, which does probe level analysis, > although not empirical > Bayes. Let me warn you though that the multiple probes for each probe set > *cannot* be treated as > simple replicates. The proble level analysis is more complex than I think > you might be expecting. > > Gordon > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor >