understanding Cuffdiff comparison (or a general understanding of pair-wise v muti-groups comparision)

Question

[CrazyB]

I have a general understanding of RNA-seq differential expression and some tools that are used to calculate the differential expression but was wondering if more information could be solicited here.

Specifically, if I have 4 different groups of samples to compare the differentially expressed genes, using Cuffdiff, are the following two approaches generating the same/different output list of differential genes -

approach1 - pair-wise comparison

[group 1 v group 2], [group 1 v group 3], [group 1 v group 4] .... [group 4 v group 2], [group 4 v group 3]

approach 2 - "all four groups" comparison

[group 1 v group 2 v group 3 v group 4].

We have actually generated all lists from both approach 1 and approach 2 and are in the process of cross-comparing them, but I was a little surprised to find that using the second approach, the output is still listed (more or less) like [group 1 v group 2], [group 3 v group 4] ...

Any references/pointers on the differences between approach 1 and 2 will be greatly appreciated. Thank you

[Ram]

It is eerie how this question exactly mirrors one of the questions I had on my mind. 4 cohorts, pairwise vs across the board comparisons - everything :)

Answer 1

[Devon Ryan]

You end up doing pair-wise comparisons either way. The difference between approach 1 and approach 2 is that with the latter you have more information available to estimate variance and will often (though not always) get a bit better results with it. I imagine your list will be similar with just some movement around the edges of the significance threshold (if not, then you likely have much more/less variance in one group).