Question: DESeq2 w. pathway/network info as a prior knowledge
2
3 months ago by
University of Copenhagen
guidobarzaghi20 wrote:

Good evening everyone,

I was wondering if anyone has ever tried to perform a differential expression analysis using pathway/network information as prior knowledge. To be more specific, I am wondering if it is possible to feed DESeq2 the prior knowledge of pathways so that the algorithm is more prone to consider statistically significant genes part of an up-/down-regulated pathway even without a big LFC (and vice-versa).

ANY help here will be highly appreciated!

deseq2 pathway ihw • 367 views
modified 12 weeks ago by Michael Love23k • written 3 months ago by guidobarzaghi20

Perhaps GSVA is what you’re looking for? It’s a Bioconductor package that shifts the differential expression analysis from genes to gene sets (e.g.: pathways).

Answer: DESeq2 w. pathway/network info as a prior knowledge
3
12 weeks ago by
Michael Love23k
United States
Michael Love23k wrote:

To answer your exact question, you could use IHW to estimate weights for partitions of the genes. It would give a better chance to those genes in a partition that has more power under an alternative hypothesis. As long as the partition is independent of the p-values under the null, which it should be.

One important note: users should not run gene set testing on a dataset, find the significant sets and then rerun with IHW using a particular partition based on the first run results. This would obviously violate the independence of the p-values and the partition under the null, but just to be very clear about that. The partition has to be decided prior to looking at the results.

However, I think a more straightforward approach is simply gene set testing. I like to use goseq following DESeq2, but other options are roast and camera from the limma package (these cannot be used directly with DESeq2).

As far as IHW is concerned, in my case the covariate used to compute the weights would be categorical (ideally a single number indicating with pathway is that gene involved in), wouldn't that be a problem? Otherwise, how else would you think of feeding that information to the model?

It’s not a problem, take a look at the paper and vignette for IHW.

I think this approach is basically just what you asked for.

We don’t have any easy built in way to give prior info to change DESeq2 test statistics.

Perfect! Thanks a lot for your help (again)