How to correct unknown Batch effect?

Question

[luffy]

Dear all,

I have RNAseq data with unknow batch information. The data has control and two disease groups. My question is since batch information is not known is there any way to deal with this?

This is a PCA plot of the data with control and 2 disease groups. Same kit and lib perp strategy has been used but no batch information

Any help would be much appreciated

Thank you

Answer 1

[Meisam]

You can perform batch effect removal with the sva R package. You can download the package and read its tutorial from its bioconcuctor page. In short, the package enables you to perform both "supervised" and "unsupervised" batch effect removal to transform your data, depending on whether the batch effected groups are known to you or not. Respectively, you can use ComBat() function and sva() function from the package to do the necessary transformation.

[luffy]

@Meisam, Thank you i will explore the sva package.

Answer 2

[ATpoint]

Typically you would first run something like PCA to see whether there is any evidence for a batch effect. If so then you can include the batch into the experimental design of for example edgeR, see Basic normalization, batch correction and visualization of RNA-seq data

[luffy]

ATpoint , Thank you, yes i have plotted the PCA using variance stabilizing transformation (VST) data from DESeq2 package. Here i do not have a batch info to include in the meta file (design file) to run the DESeqDataSetFromMatrix function with ~ group + batch.

[ATpoint]

You can estimate unknown sources of variation with svaseq (sva package) or ruvseq. There seems to be at separation in PC2 irrespective of group. Go through mentioned functions from the packages and see whether it makes sense to remove some of the signal not separating groups but other signals.