Is there any example for Differential gene expression ?

Question

[Mo]

I am desperate of doing DGA on my data. So far, I could not find any guide which helps me to perform it and I could not find a reason why I cannot do it. Here I write my idea maybe one who really knows the technique can guide me to perform it.

I have a Matrix-1 (each row is a gene and each column is a sample) this matrix is controlled

I have another matrix-2 ( each row is the same gene as matri-1 and each column is a sample) but this one is untreated.

Now, I want to find those genes which are unregulated and those which are down regulated. There are over 100 comments, 100 packages to perform DA. some says Limma, Dseq and so many other packages, some says pairwise T-test , multi comparison. There is an any well-written comment or guide showing what we need and why we cannot perform such analysis using my data.

By the way, I also tried to filter out those genes which do not express by genefilter method as follows:

f1 <- pOverA(0.25, 3.5)
ffun1 <- filterfun(f1)
flrGene <- genefilter(data,ffun1)
sum(flrGene)

Then it gives me zero, why? Means I should keep all the genes? Is there any other method to remove those genes with very low expression over samples?

Answer 1

[Devon Ryan]

1. For a microarray, limma is the package of choice. You would never use DESeq (presumably this is what you mean by "Dseq") for microarray data. Yes, you can perform a T-test, but this will have lower power (see the limma paper).
2. The limma user guide is extremely thorough and well written. Read it. BTW, you'll need your data in a single matrix for essentially any package.
3. Without knowing the details of your dataset (i.e., looking at the actual numbers), no one can tell you why that filter command is producing all false values. Presumably 3.5 is inappropriate for your dataset. Look at a few rows to see whether the results make sense.

[Mo]

@Devon Ryan thanks for your comment but I read some many posts which all say use Limma, none provided any information, how to do so!

I found where the problem was with filtering. but do you think it is a good approach to remove genes which are not highly expressed? Is there any other technique?

[Devon Ryan]

Read the limma user guide, it provides all the information you're likely to need and has a lot of examples.

Regarding removing lowly expressing genes, yes this is a good approach. See the Bourgon et al. 2010 PNAS paper referenced from the genefilter tutorial for why. I have yet to see a better method presented than this. BTW, part of the point of that paper is that you don't need to choose an arbitrary threshold, but can find a threshold that maximizes your statistical power. Again, read the paper the package vignette.

[Mo]

@Devon Ryan I have tried to do so, although I could not find an exact example similar to my data structure

r <- romer(fulldata=fulldata,index=list(set1=index1,set2=index2), design=design,contrast=2,nrot=99)

fulldata is the matrix of expression values (each column is a gene and each row is a sample)

index, in fact, I don't know what it is but I just put the same as an example

design = I made a two column matrix corresponding to the control and untreated samples

contrast, I could not find how to tune this but I mentioned 2 as classes

nrot=99 which corresponds to number of rotations used to estimate the p-values

I have got an error already as follows:

Error in as.matrix(y) : argument "y" is missing, with no default

Answer 2

[gmesserli]

Dear Mo,

If you are still looking for a solution to easily analysis your gene expression experiment, I could suggest you to visit the following page: https://genevestigator.com/gv/file/GENEVESTIGATOR_private_data_service.pdf

Beside a very simple and user-friendly tool, GENEVESTIGATOR offers the advantage of having your own experiment "privately" integrated to thousands of well annotated and curated other experiments so that you can see your results in a much broader context and analyze them in novel ways.

Good luck