Question: Limma log transform RNA-seq and microarray
0
3.4 years ago by
bharata180340
Japan
bharata180340 wrote:

Hello,

I'm working with Limma for both RNA-seq data and microarray data. I used gene read count for RNA-seq data as input for Limma. As for microarray data, I use normal workflow for Agilent microarray data.

So, I just want to ask how can I get a matrix of log transform of RNA-seq values and microarray intensities using Limma?

I expect the result would be like a matrix with columns are the name of the sample and the rows are the genes ID/name. I know for DESeq2, there is function to do that but I don't know for Limma to do the same thing. Thank you for your answers.

limma • 1.6k views
modified 3.4 years ago by Aaron Lun23k • written 3.4 years ago by bharata180340

When you say you are using limma for rna-seq, I hope you mean that you are using voom?

Yeah, I mean like that.

Answer: Limma log transform RNA-seq and microarray
2
3.4 years ago by
Aaron Lun23k
Cambridge, United Kingdom
Aaron Lun23k wrote:

Well, it depends on what you want to do with the log-transformed values. If you just want to use them for visualization, then you can use the cpm function with log=TRUE in edgeR. This will compute a matrix of log-counts per million for each gene in each sample. If you want to do DE testing, then you should use the voom function. This will also compute log-CPMs but with additional precision weights to account for the change in the mean-variance relationship between small and large counts. These log-values and their weights can then be used in the standard limma pipeline, e.g., lmFit, eBayes, and so on.

For the microarray intensities, it should be easy enough to extract log-transformed intensities from the object you get after normalization, usually as object\$E. However, it should be stressed that you really shouldn't analyze microarray and RNA-seq data in the same model. The technologies are just too different, and I don't think it'll make any sense to try to combine information between them for variance estimation. A better approach would be to perform separate analyses and compare the final DE lists. Then you can see whether the biology in the two experiments is consistent or not.