RNA-seq: TPM batch effect correction

Question

[Samuel]

Hi all,

I conducted RNA-seq on S. cerevisiae strain grown by triplicates in two conditions agave juice and minimal media.

My goal is to compare gene expression between two genes within the same condition. For that purpose, I was thinking to perform bacth correction with Combat-seq and then apply TPM normalization for gene-wise comparison. However, I realize this approach may not be optimal. ComBat-Seq is designed to work on raw counts assuming a negative binomial distribution, whereas TPM is a derived quantity that alters the scale and distribution of the data. Applying TPM after ComBat-Seq could distort the batch correction.

I would really appreciate any guidance on this matter. Take into account that batch correction is needed since the replicates for the same condition where extracted on different days (batches)

Answer 1

[rfran010]

I don't think batch correction is necessary here, especially since you just want to compare gene expression within a condition. As long as there's no explicit difference in the replicates, meaning logically you expect the data is consistent, then I would say the variation from collecting different days is good variation.

Answer 2

[jared.andrews07]

I'd consider taking a simpler approach and just looking at gene ranks by TPM first. Do you know that the batch effect affects ranks in a meaningful way?

Look at robust rank aggregation as a way to aggregate ranks within each group.

Answer 3

[i.sudbery]

I'm with @rfran010 here. If you're goal is to compare genes within samples, then since all measurements from within a sample come from the same batch, there is no need to batch correct. By comparing two genes within a sample, you are effectively doing a paired test.