Hi all,
I have RNASeq data for 4 different conditions (si41, si42 and si43 as siRNA's for my target and then a siNTC as control) and each condition has 4 reps. If I cluster them in a PCA plot, there is clear batch effect for replicates, meaning all samples for replicate 1 cluster together, all samples for replicate 2 cluster together and so on. There is no clear separation of the samples based on the conditions. This of course resulted in very few differential expressed genes (I use Limma/Voom).
I tried using Combat to remove this batch effect which was quite successful. My samples now cluster together based on their condition but my PC1 and PC2 are only 11%. I'm now stuck with how to move on to differential gene expression analysis.
I also already tried to use "removeBatchEffect" form the Limma/Voom pipeline but this does not really remove all of the batch effects.
Does anyone know how to move on?
Did you take replicates at the same time, or different times? Regardless, it might be that the effects of your treatments are relatively small. Alternatively, the effect of each treatment could have been nested within batch effect (i.e., magnitude of effect was larger when all rep 4s were measured compared to rep 1s), so you removed it when removing batch effect.
Hi, yes indeed we took all the reps at a different time. And I totally agree with you that the effect of the treatment is probability too small, however the top down regulated target is indeed our target that we treated so I would just expect that other genes would move along.
Have you any idea on how to do DE analysis on the expression matrix after combat?
The targeted gene is downregulated. You did everything at the same time and replicate clustering is stronger than treatment clustering. Sounds to me that the knockdown has no transcriptional effect. No DEGs is not necessarily a bad thing or technical issue. It can simply mean that there is no effect.
In addition to biological explanation, maybe the data is still too noisy, so DEGs aren't called? What do the logFCs look like?
I don't know how combat works, but have you tried to run DEGs with a paired setup (where siRNAs are compared to NTC within each replicate)?
Hi, the logFC's are quite large between 5 and -6 which are pretty nice values but it's just not significant.
Can you help me with the code for a paired setup? Many thanks!
See the vignette of the tool you use. Basically it's
~pair+condition. LogFCs alone mean nothing, as it could be large with even larger standard errors, due to lots of noise.