p values, q values, stat.mean for gage() outputs - What to report and why?

Question

[jpvoth]

Hi everyone - I am working on some differential expression and pathway analysis and am finding it difficult to determine what to report in figures for this. The literature on this is pretty inconsistent so I figured I would query you all.

When I run gage(), I get outputs that include a p-value, q-value (adjusted p value for false discovery rate I presume), and stat.mean. It seems stat.mean is the mean log2-fold change across the gene set being assessed by default according to the gage documentation. This seems like the most important metric to report, as it indicates whether the pathway itself is up or down as a whole, right? Typically, I would assume the q-value is the more important than p-value for this type of analysis since it is adjusting for multiple tests. However, the p-values and q-values are pretty drastically different, while the stat.mean is still pretty substantial. So - I am questioning whether to consider/report the p-values or q-values for these analyses. I have pasted a chunk of an output as a sample to show the wide range between p.val and q.val for your information.

I'm working through this and am not really asking for code help, rather interpretation and insights into best practices. Thanks a lot for your time here.

Answer 1

[BioinfGuru]

I would only include q.val and set.size. In the image you can see very similar q.val for highly different set.size. For a cut off of q.val <0.05 the set.size will help ranking

Possibly stat.mean. - "[stat.mean] measures the magnitude of gene-set level changes, and its sign indicates direction of the changes". I'm not convinced that stat.mean is adding anything useful to the story, What if some of the gene set is up regulated and the rest are down-regulated - do they cancel each other? What does that do to stat.mean? Maybe the extreme stat.mean values are strong results, but surely that would be reflected in the q.val anyway.

Lots of papers report lots of stats just because they have the stats and forget to ask the question, "does this add noise, or is it important in telling the story"? You know the data, does stat.mean tell you anything you don't already know?