DESeq2 results difference using contrast and single conditions

Question

[User000]

Dear all,

I have 4 conditions and I want to find up/down regulated genes using DESeq2 by comparing condition1 vs condition2, condition3 vs conditions4, and condition1 vs conditions3. I tried two methods:

1. I uploaded all conditions with all 3 replicates and created one single table with normalised raw counts. I used contrast in order to get the comparison between the desired conditions:

    dds <- DESeq(dds)
    res <- results(dds)
    res <- results(dds, contrast=c("condition","condition1","condition2"))
    resultsNames(dds)
    contrast=c("condition","condition1","condition2")
    resLFC <- lfcShrink(dds, alpha=0.001,  lfcThreshold = 1, contrast=contrast, type="ashr")
    resLFC
    up <- subset(resLFC, log2FoldChange > 1)
    down <- subset(resLFC, log2FoldChange < -1)

1. I uploaded raw counts only for conditon1 and condition2, normalised and performed DESeq2 and used the same contrast option above.

My question is, why am I getting different results (i.e. number of up/down regulated genes)? Which is the right/best way? Finally, how should I treat the biological replicates (merge them (if yes,when?) or keep them as separate samples).

Answer 1

[bigomics.team]

Most probably because of the estimation of the variance. DEseq2 uses a Bayesian method. In the integrated approach the variance is estimated using more samples. Check the foldchange, they should be similar. There is no correct way. Just see which one gives you more significant values and describe the methods you choose. The "integrated" approach is less work.

Answer 2

[swbarnes2]

In general, include more samples unless you have a reason not to; you'll get better size normalization and dispersion estimates

See this as an example of when you should split up your data:

http://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#if-i-have-multiple-groups-should-i-run-all-together-or-split-into-pairs-of-groups