Removing confounding batch effects

Question

[Kaiwen]

Doing DEG analysis on disease vs. control and I have RNA-seq data from multiple sources. The problem is that a few sources have only disease or only control samples. Already tried DESeq2's built-in method of adding the batch covariate but it didn't work too well, as it was showing multiple genes with |lfc| > 15. Combat also didn't work great.

Looking for any other suggestions. Not sure if it is still mathematically possible to remove the batch effects when they are confounded with the condition effect.

[ATpoint]

when they are confounded with the condition effect

It's not. DESeq2 and limma all follow the same underlying principles. There is no magic to separate batch and condition if nested.

[SamGG]

Hi. Show clearly your design.

[Kaiwen]

Something like this: