Can we directly use batch corrected gene expression (Bulk RNA-seq) for further downstream analyses?

Question

[tigeradab]

Although I understand that removal of batch effects can help with estimating the actual variance in the gene expression data, I would like to know whether the batch corrected gene expression values can be directly used for finding transcriptionally abundant genes across experimental studies?

Answer 1

[Kristoffer Vitting-Seerup]

You need to distinguish between two types of downstream analysis:

1. Analysis which itself uses a (linear) model.
2. Analsis which "just" uses the expression values.

1) Modeling analysis

For analysis which itself uses a model - such as differential expression - you should provide the raw (count) data and build the batch effect into the design matrix you use with that model. Doing this is easy with the standard DE tools (edgeR, DESeq2, Limma) and will result a batch corrected DE analysis.

2) Non-modeling analysis

This is for analysis such as unsupervised clustering, heatmaps, PCA etc. Here you need to provide batch corrected data as there (in standard implementation) are no way of removing batch effects.

Hope this helps

Cheers

Kristoffer

[tigeradab]

Thanks for the answer. Very useful for a novice like me. To specify further, I don't want to perform a differential analysis between control and treated samples across studies. I want to take the control of different studies and perform sort of a meta-analysis. In this context, I would like to know whether I can use the batch-corrected average expression of genes to determine a threshold to classify transcriptionally abundant genes.