Batch effect that is same as main factor

Question

[joonhong kwon]

Hi all,

I have RNA-seq data of 54 samples from 6 mouse brain regions with 3 different groups. One group is control (no stress), another group is stress-resilient (Res) and the other group is stress-susceptible (Sus).

These 54 samples are sequenced in 5 different days. Most of the data with the same brain region are sequenced at the same day. (RNA extraction was performed on the same day for all samples. And library prep. and sequencing was performed by region. That is, sequencing and prep. date will be the same.)

So I tried to perform batch correction using ComBat, but it seems that the difference by region was normalized.

In this case, any suggestion?

This is my ComBat R script:

datexpr <- read.csv("FPKM.csv", row.names=1)
trait <- read.csv("trait.csv")
trait$seqdate<-as.factor(trait$seqdate)
batch <- trait$seqdate

datExpr.combat = ComBat(dat=as.matrix(datexpr), batch=batch, mod=NULL)
write.csv(datExpr.combat, "Corrected_FPKM.csv")

And my data information (total 54 samples):

    Region State Seqdate
1    AMY   Con   191214   x 3 (ex. AMY_Con1, AMY_Con2, AMY_Con3, same below)
2    AMY   Res   191214
3    AMY   Sus   191214
4    HIP   Con   191214
5    HIP   Res   191214
6    HIP   Sus   191214
7    CBC   Con   190826
8    CBC   Res   190826
9    CBC   Sus   190826
10   NAc   Con   191029
11   NAc   Res   191029
12   NAc   Sus   191029
13   PFC   Con   200317
14   PFC   Res   200317
15   PFC   Sus   200317
16   VTA   Con   200427
17   VTA   Res   200427
18   VTA   Sus   200427

Refer to the following question.

Batch correction using DESeq2

Thank you!

Answer 1

[swbarnes2]

You could correct for library prep date, or possibly RNA extraction date, but you don't need to correct for sequencing date. Which is good for you, because you totally confounded sequencing date with region.

[joonhong kwon]

oh sorry,

Apologies about the confusion. I edited my post.

RNA extraction was performed on the same day for all samples. And library prep. and sequencing was performed by region. That is, sequencing and prep. date will be the same.

So, I can't correct batch effect.

[swbarnes2]

If you prepped all the samples of one tissue type on one day, and another tissue type another day, you screwed up, and math can't fix that. Do the intra-tissue comparisons you can, and next time, plan better.

[joonhong kwon]

Thank you, I will have to consider intra-tissue comparisons.

[Zhenyu Zhang]

As said by swbames2, you confounded region with date in your experimental design, so now every time you try to correct for batch effects in date, region will also be corrected. You have 3 choices 1) re-design your experiement 2) don't do batch correction 3) figure out a complicated stat model to separate effects from date and region if they are not 100% confounded. but please remember every reviewer will question your method.

[joonhong kwon]

Thank you for your reply!