Is it possible to use edgeR and Limma for meta analysis?

Question

[Pas]

Hi all,

I was wondering if it is possible to use edgeR and Limma to perform meta-analysis of gene expression data generated from different platforms. In particular I'd use the generalized linear model GLM and a design model where I can adjust for differences between the different platforms by using an additive model formula like this:

design <- model.matrix(~platform+Treatment)

What do you think?

Answer 1

[andrew.j.skelton73]

While this is possible, there's a lot of caveats to it... You'll need the same sample types on both platforms (to estimate within platform variance). If you're using microarrays, you'll have to match probes that are the same up (note, that these would have to be pretty damn close, i.e. mapped up nuIDs so that they have the same nucleotide sequences) - There are packages like inSilicoMerging. Either way, you generally need to take this with a pinch of salt.

If you're using doing this for RNA Seq, I'd recommend using DESeq2 for gene level and Sleuth for Transcript level - Both can handle additive models. If you're using microarrays, use Limma.

[Pas]

Thank you Andrew.

I agree with you. I heard that people are trying to use egdeR or limma to perform a meta analysis integrating both microarray and RNAseq, specifying the platform in the design model, and since I 'm not sure that is possible, I'd like to have more opinions.

I have many concerns about this approach. For example, what about the normalization? Microarray data cannot be normalized like RNASeq data, and edgeR needs raw data (not normalized) as input. Then, giving to edgeR a matrix containing raw data from both RNAseq and microarray it'd be a error. Do you agree?

[andrew.j.skelton73]

Woah, while I'm saying it's possible with the same technology (i.e. 2 microarrays combined, or two RNA Seq datasets combined), I have to highly recommend against combining 2 different technologies (i.e. microarray and RNA Seq dataset). There are ways to look for correlations between microarrays and RNA Seq data, but combining for differential expression analyses and such would be more trouble than it's worth going down that rabbit hole.

[ma_nezami65]

Thank you Skelton,

I have 5 RNA-Seq datasets and I want to perform meta analysis. the batch effect removing was done by ComBat and it gave continuous data. you wrote that, we can use DESeq2 or limma to perform meta analysis?

You mean we can consider all control samples or all treatment samples in 5 studies as replications? honestly, I have a problem with my two datasets because they don't have replication and i can not use package like metaDE or any other packages. on the other hand, due to continuous data resulted from ComBat I can not use metaSeq.

Have you read an article which it used deseq2 or limma for meta analysis?

Thanks