multifactor microarray data sample analysis using limma package

Question

[jaichandpatel2]

I had microarray data sample files (.CEL files). I had used limma package & found out topTable of differentially expressed genes in R . But now I have some samples in which more than one factors have been applied i.e. strain, time, treatment etc. So I am not able to construct design matrix. Although in limma user guide has given an example of estrogen datasets but I didn't get it.

Please help me if anyone have any idea regarding object of experimental design. How can we add factor with raw data i.e. .CEL files. please give the answer with any example if you can.

[Devon Ryan]

What part of that user guide didn't you get exactly?

[jaichandpatel2]

9.5 Interaction Models: 2 ? 2 Factorial Designs

Can you help me to understand above part?

[Devon Ryan]

Yes, what part of that section don't you understand?

[jaichandpatel2]

I have a data GSE35819. I want to analyse this data.

In this data I have 18 samples (18 .CEL files)

3 cell line, 3 time points. can you tell me how to design matrix, contrast matrix.

I have some more data of this kind.

plz help as much you can.

Answer 1

[Devon Ryan]

The general idea with GSE35819 is:

design <- data.frame(cellLine=factor(c(rep("H9",3), rep("HS401",6), rep("HS360",6), rep("H9",3))),
    timePoint = factor(c(24,7,7,2,2,24,24,7,7,2,2,24,24,7,7,2,2,24)),
    treatment=factor(c(rep(c("Normoxia","Hypoxia"), 12)))

The model matrix is then dependent on what you want to test. Perhaps you just care about "treatment" while controlling for everything else (excluding interactions):

mm <- model.matrix(~cellLine+timePoint+treatment, design)

You don't always need contrast matrices, it depends on the model and the question. In general, you should either take a class on linear models or just talk to a local statistician about individual designs. Providing a general tutorial on linear model is well beyond the scope of this site (that's why there are classes and books dedicated to the subject).

[jaichandpatel2]

Thank you sir.

[jaichandpatel2]

Can I use maSigPro package for this. As I used maSigPro package and got this error. What is meaning of this error, should I change samething in this R-script.

If you have any idea then suggest me.

> edesign
        Time Replicate Control hypoxic
Array1     3         1       1       2
Array2     2         1       2       1
Array3     2         1       1       2
Array4     1         2       2       1
Array5     1         2       1       2
Array6     3         2       2       1
Array7     3         2       1       2
Array8     2         2       2       1
Array9     2         2       1       2
Array10    1         3       2       1
Array11    1         3       1       2
Array12    3         3       2       1
Array13    3         3       1       2
Array14    2         3       2       1
Array15    2         3       1       2
Array16    1         1       2       1
Array17    1         1       1       2
Array18    3         1       2       1

fit <- p.vector(eset, design, Q = 0.05, MT.adjust = "BH", min.obs = 20)
Error in dat[, as.character(rownames(dis))] : subscript out of bounds

[Devon Ryan]

I know nothing about the maSigPro package.

[jaichandpatel2]

Hello Sir

I have got p-adjusted value i.e. q-value using limma package, but how can I get the gene list of different coefficients in an ascending or descending as well in .xls format. what should be the cut off in case of q-value to select differentially expressed genes (up-regulated & down-regulated ), Will it be same as in p-value (0.05 or 0.01) or something else? Please Sir suggest me, what should I do ?

Thank you very much

[Bioinformatist Newbie]

Hi Devon,

Kindly shed some light on following post, thanks.

Microarray analyis of different platform (GPL96 vs. GPL3921)