Question

technical replicates and lmFit error

0

Entering edit mode

Ommen Kloeke, A.E.E. van ▴ 10

@ommen-kloeke-aee-van-5409

Last seen 9.6 years ago

Dear Bioconductor team, I am working on a 2-colour Agilent dataset which has both biological and technical replicates. A problem which many people already encountered as I could see from the previous posts, so my apologies for bringing it up again. However, I tried the script given for this situation in the Limma user guide, but my lmFit gives an error if I try this! My experiment contains three treatments: "AC", "low" and "high" - each has 4 biological replicates and 2 technical replicates as indicated in the attached target file. The main contrasts of interest are "low-AC" and "high-AC". Here's the script I tried: design = modelMatrix(targets, ref = "AC1") design = cbind(Dye = 1, design) colnames(design) #[1] "Dye" "AC2" "AC4" "AC5" "high1" "high2" "high4" "high5" "low1" "low3" "low4" "low5" fit = lmFit(MAbet, design) cont.matrix = makeContrasts(ACvsLow = (high1+high2+high4+low5-AC2-AC4-AC5)/4,levels = design) fit2 = contrasts.fit(fit, cont.matrix) fit2 = eBayes(fit2) topTable(fit2, adjust = "fdr") However the LmFit gives an error: > fit = lmFit(MAbet, design) Coefficients not estimable: low4 low5 Warning message: Partial NA coefficients for 43803 probe(s) I understand this has to do with my design, but I don't know how to fix it: > design Dye AC2 AC4 AC5 high1 high2 high4 high5 low1 low3 low4 low5 [1,] 1 0 0 0 0 0 0 0 0 0 1 0 [2,] 1 -1 0 0 0 0 0 0 0 0 0 1 [3,] 1 0 1 0 0 0 0 0 -1 0 0 0 [4,] 1 0 0 1 0 0 0 0 0 -1 0 0 [5,] 1 0 0 0 0 0 1 0 0 0 0 0 [6,] 1 -1 0 0 0 0 0 1 0 0 0 0 [7,] 1 0 1 0 -1 0 0 0 0 0 0 0 [8,] 1 0 0 1 0 -1 0 0 0 0 0 0 [9,] 1 0 0 0 -1 0 0 0 0 0 1 0 [10,] 1 0 0 0 0 -1 0 0 0 0 0 1 [11,] 1 0 0 0 0 0 1 0 -1 0 0 0 [12,] 1 0 0 0 0 0 0 1 0 -1 0 0 > is.fullrank(design) [1] FALSE I completely trust in your expertise! Any help is very welcome and appreciated. Much obliged and many thanks! Elaine van Ommen Kloeke VU university Amsterdam Department of Ecological Science room: H-119 phone: 020-5987217 www.falw.vu.nl/animalecology<http: www.falw.vu.nl="" animalecology=""> -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: targetsE2phen.txt URL: <https: stat.ethz.ch="" pipermail="" bioconductor="" attachments="" 20120719="" 36025cdf="" attachment.txt="">

limma limma • 727 views

ADD COMMENT • link updated 23 months ago by Gordon Smyth 50k • written 11.8 years ago by Ommen Kloeke, A.E.E. van ▴ 10

score 0 · Answer 1 · 2012-07-21

Dear Elaine,

Well, you are trying to fit a model with 12 coefficients to a dataset with only 12 arrays. It is no surprise that this causes a problem.

I have had a look at the targets file. If I understand the meaning of the labels is the file, it appears that the mixing up of technical replicates with biological replicates is very complex. So complex that I do not know of any practical approach that would unravel them in an analysis.

I suggest that you simply treat all your technical replicates as ordinary biological replicates, so that a ordinary limma analysis is straightforward. Then take the p-values that you get at the end with a grain of salt, because they will be somewhat smaller than they should be.

Best wishes
Gordon

PS. lmFit has given a warning, not an error. There's a difference!