meta-analysis of RNA_seq public data

Question

[l.nazari]

I would like to do meta-analysis on the public data of RNA_seq downloaded from several studies. which packages do you recommend for differential expression? Can I use meta DE that is specific for microarray? I have already downloaded the data and have done quality control by CLC. Thanks in advance

[ATpoint]

I personally do not know the package you mention but one package I tried is here in this comment from earlier today C: Normalization method to be used when dealing with multiple datasets

[b.mascat]

That's edgeR sorry for the typo. You can introduce the batch effect as a co-variable in the experimental design in DEseq2 or correct it with limma as far as I know.

[GenoMax]

You can edit your posts by choosing edit link you see at bottom of your posts/comments. Submit Answer should be used only for new answers to the original question.

[ATpoint]

You can model batch effects as part of the design, that is true but as said that requires replicates of each condition in each batch. If you have like all of condition-A in study-A and all condition-B in study-B then it is perfectly confounded and you cannot distinguish condition effects from batch effects. Be aware that both DESeq2 and edgeR are model-based frameworks so they take raw counts, therefore you cannot feed in corrected counts directly. There are many threads both here at biostars and over at support.bioconductor.org on that topic if you would like to dive into it.

Answer 1

[b.mascat]

If all your datasets were sequencing by RNA-seq you can use some R package like DEseq2 or Edje2 but you have to take care about how this rna-seq was performing (Bulk RNA, PolyA-RNA, rRNA deplection) and the technology that was used (Ion torrent, Illumina, etc). Depending of this, probably the pre-procesing of the datesets won't be the same.

[ATpoint]

You probably mean edgeR, but this is likely not going to work out reliably as you cannot simply ignore the influence of batch effects. Meta-analysis sounds about right here unless you have replicates of all groups you want to compare in all of the studies. If not you your analysis is confounded.