Hi all,

First off, I'm a wet lab scientist learning to analyse my own data. I've design my experiment as such

> coldata
    sample        condition litter
1 KO1_sort.bam        KO      A
2 KO2_sort.bam        KO      B
3 KO3_sort.bam        KO      A
4 KO4_sort.bam        KO      B
5 WT1_sort.bam        WT      A
6 WT2_sort.bam        WT      B
7 WT3_sort.bam        WT      A
8 WT4_sort.bam        WT      B

Before normalization, WT3 and KO1 shows higher variability from the RLE plot, and also cluster together based on the first principal component on my PCA plot.

RUVg with k=2 is able to reduce the variation seen in the RLE plot and results in WT and KO samples clustering separately on PCA plot.

Empirical genes for RUVg were obtained using a cutoff of pvalue > 0.5 and design = ~ litter + condition in DESeq2. My question is whether I should still account for the 'litter' factor in my DESeq2 design after taking into account the variation modelled using RUVg, or not? Option 1:

design(ddsruv) <- ~W1 + W2 + litter + condition

Option 2:

design(ddsruv) <- ~W1 + W2 + condition

Thank you!

I would opt for design 1 here. There's not much harm to adding a simple coefficient that mark litter, and I think given how you ran RUVg W1 and W2 should be somewhat orthogonal to litter.