How do I remove Batch Effects from Two Separate raw RNA SEQ datasets

Question

[merfer0206]

Hello there,

I have a question regarding to account for batch differences in RNA-seq data analysis.

Originally, an RNA SEQ data set was sequenced with two conditions (Day 0, Day 30), and then a couple months later we sequenced another data set (Day 30, Day 90). I now want to analyse the full data set (Day 0, Day 30, and Day 90). However, I am having trouble understanding how to account for batch differences. When I look at the raw dataset, the main issue to me is that my two data sets contain different row numbers (batch 1 = 583400, batch2= 63682). I was told to use ComBat-SEQ, but I am really lost with this.

If anybody could point me in the right direction I'd be really grateful!

Answer 1

[rpolicastro]

Batch correction in DESeq2 or edgeR would incidentally work because you have samples from Day30 in both batches. In your DESeq2 or edgeR regression model add batch as a covariate to your formula

~ condition + batch

[merfer0206]

Coming in with another stupid question, as I have two sets of Day 30 of triplicate repeats, so how do I navigate around that?

[rpolicastro]

Are all the Day30 samples independent biological replicates? If so, just treat them as separate samples.

Answer 2

[ATpoint]

I would encourage you though to first do exploratory analysis via PCA (see DESeq2 manual for this) and see whether there is evidence for a batch effect, so do day30 samples cluster well together, or is there an obvious shift between samples from the different batches. Only if there is evidence for the batch effect go ahead correct it as rpolicastro suggests, and if there is no evidence then do not correct it.