batch effect after using comBat

Question

[Inbalor]

Hi!

I am analyzing data from RNA-seq. I have six groups: group1 (batch1), group2 (batch 1 + batch2), group3 (batch2), group4 (batch1), group5 (batch2) and group6 (batch1+batch2). My main interest is in detecting differential expression genes between groups using limma R package. Unfortunately, my batch effect is confounded with the biological groups.

I have a strong batch effect in my data. I have tried to eliminate it using comBat. However, when I plot batch corrected data using PCA/heatmap the set effect is still there although it has been reduced.

Do you have any ideas how to diluted more the batch effect? I am detecting differential expressed genes but I think that most of them can be just because of the batch effect.

Thanks a lot for your help in advance!

[Inbalor]

I though to include the batch effect in my model to test differential expression using gene expression corrected batch data. What do you think?

[Neilfws]

You may not be able to reduce the effect any further, depending on the degree of confounding. In the worst case scenario (I've seen it!) phenotype and batch may be completely confounded; in that case, nothing can be done.

[Inbalor]

Thanks for your reply. Then, I have to assume that all the variance associated to my batch effect that I could eliminate has already been eliminated by Combat. Thanks!