Using SVA to correct for covariates in RNA-seq data

Question

[sbrown669]

I am working with RNA-seq data, and I need to correct for covariates. To do this, I want to use the Bioconductor package SVA. I want to produce a 'corrected dataset' and then use this for further analyses - to perform pairwise Pearson correlations and generate heat maps to visualise co-expression. However, as I understand it, the documentation explains how to feed the covariates into a model for differential expression analysis, not how to produce and export a corrected dataset based on the SVA calculations.

I'd appreciate any advice on doing this, or in fact being informed that I'm going about this in the wrong way.

Answer 1

[Kevin Blighe]

Hey,

Yes, from what I understand, SVA will just determine the batch covariates that [may] exist in your data. You can then use removeBatchEffect if you want to directly remove these batch effects from your data.

Be cautious of these batch detection methods, though - one does not want to remove genuine biological effects of interest that exist in the samples.

Kevin

Edit May 8, 2020:

• removeBatchEffect() only has certain use-cases, though - see Trying to understand the maths behind one Limma function (it cannot be used on raw counts or raw data, generally).
• ComBat-seq is now also a thing, and should be explored: https://github.com/zhangyuqing/ComBat-seq