Question: Agi4x44Preprocess/Limma and number of significant p-values
0
gravatar for Paulo Nuin
8.0 years ago by
Paulo Nuin200
Canada
Paulo Nuin200 wrote:
Hi everyone I have used Agi4x44Preprocess (script below) to analyze three mouse (mgug4122a) Agilent microarray sets and I want to check if the methods I'm using make sense, as the number of significant p-values I'm getting seems quite high. I'm using Limma for the significance analysis. the lowest value I'm getting is 1.09667003222346e-12, while the adjusted p-value 2.43208513046197e-08. If I consider significant p-value below 0.01, I get 9721 significant ones, roughly half of the gene list. This was for one of the datasets, with 2 (four result sets) micorarrays, 1 for each treatment. For another similar dataset, 4 microarrays with 2 samples for each treatment, results are very similar. I also tried the approach in this page http://matticklab.com/index.php ?title=Single_channel_analysis_of_Agilent_microarray_data_with_Limma and the results seem better (I mean less p-values below 0.01). I was wondering what would be the suggested approach and if there's any reason that I'm getting so many values below 0.01, with some extreme values at 10-12 range. Is there a bug on Agi4x44Preprocess? It seism to get the same columns as the approach on the MAttick's lab webpage. Thanks in advance for any help Paulo library(Agi4x44PreProcess) library(mgug4122a.db) targets=read.targets(infile="targets.txt") dd=read.AgilentFE(targets, makePLOT=FALSE) CV.rep.probes(dd, "mgug4122a.db", foreground = "MeanSignal", raw.data = TRUE, writeR = T, targets) genes.rpt.agi(dd, "mgug4122a.db", raw.data = TRUE, WRITE.html = T, REPORT = T) ddNORM = BGandNorm(dd, BGmethod = "half", NORMmethod = "quantile", foreground = "MeanSignal", background = "BGMedianSignal", offset = 50, makePLOTpre = FALSE, makePLOTpost = F) ddFILT = filter.probes(ddNORM, control = TRUE, wellaboveBG = TRUE, isfound = TRUE, wellaboveNEG = TRUE, sat = TRUE, PopnOL = TRUE, NonUnifOL = T, nas = TRUE, limWellAbove = 75, limISF = 75, limNEG = 75, limSAT = 75, limPopnOL = 75, limNonUnifOL = 75, limNAS = 100, makePLOT = F, annotation.package = "mgug4122a.db", flag.counts = T, targets) ddPROC = summarize.probe(ddFILT, makePLOT = F, targets) esetPROC = build.eset(ddPROC, targets, makePLOT = F, annotation.package = "mgug4122a.db") write.eset(esetPROC,ddPROC,"mgug4122a.db",targets) levels.treatment = levels(factor(targets$Treatment)) treatment = factor(targets$Treatment, levels = levels.treatment) design = model.matrix(~0 + treatment) print(design) colnames(design) = c("EV50", "SHGR19") fit = lmFit(esetPROC, design) CM = makeContrasts(EV50vsSHGR19=EV50-SHGR19, levels=design ) fit2 = contrasts.fit(fit, CM) fit2 = eBayes(fit2) my.lfc <- 0 output <- topTable(fit2, coef=1, adjust.method="BH", lfc=my.lfc, number = 20000) write.table(output, file="output.txt", sep="\t", quote=FALSE)
ADD COMMENTlink modified 8.0 years ago by Jarek Bryk110 • written 8.0 years ago by Paulo Nuin200
Answer: Agi4x44Preprocess/Limma and number of significant p-values
0
gravatar for Jarek Bryk
8.0 years ago by
Jarek Bryk110
Jarek Bryk110 wrote:
Hi, > This was for one of the datasets, with 2 (four result sets) micorarrays, 1 for each treatment. For another similar dataset, 4 microarrays with 2 samples for each treatment, results are very similar. Could you tell me again how many samples you had per group to compare? And more details about the experimental design. For example, if you grouped your samples so that you had your experimental groups on separate slides, then some technical variation could have contributed to the differences. If one of your groups had some very severe treatment (or if samples were labeled in a different batch, or if you compare wild vs inbred mice) that could also be the reason. 50% of differentially expressed seem high at first sight. I am not sure, but if it's real, the quantile normalization may break for such a divergent set (others will weigh in on that). Also, I didn't get if you talking about genes or probes, and whether you did some non-specific filtering of your data. cheers jarek -- ?Jarek Bryk | www.evolbio.mpg.de/~bryk ?Max Planck Institute for Evolutionary Biology ?August Thienemann Str. 2 | 24306 Pl?n, Germany ?tel. +49 4522 763 287 | bryk at evolbio.mpg.de
ADD COMMENTlink written 8.0 years ago by Jarek Bryk110
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 389 users visited in the last hour