how to do Batch effect

Question

[elhamdallalbashi]

I have RNA-seq data,I did differential experssion,now I want to performe meta-analysis.I have 3replicates control,3 replicates drug A,3 replicates drug B and 3 replicates drug C. What would its batch variable be?

Answer 1

[Persistent LABS]

I am not sure how metaDE will be used to remove batch effects. Another option is Combat (http://www.bu.edu/jlab/wp-assets/ComBat/Usage.html). Here is a review article that compares various batch-effect tools in RNA-Seq (http://www.tran-med.com/article/2016/2411-2917-2-1-3.html#outline_anchor_5). You might find it useful.

Answer 2

[andrew.j.skelton73]

@Persistent LABS - Methods such as ComBat have been shown to exaggerate confidence when retaining group differences, so it's probably not the best way to go about things (see here).

@elhamdallalbashi - You haven't really alluded to what you're analysing, I'm assuming that you're trying to do a meta analysis between two RNA seq datasets, then you really need to consider if meta analysis is appropriate. To "combine" the data, then you need sample groups to be common across both experiments, this allows you to measure what is technical variance and what's biological signal. If you're unable to design an appropriate model that encapsulates the differences between the datasets, then you could consider non-parametric approaches like rankprod, or even heading towards a bit of set theory; intersections and set differences.