Minimal Req'D Info For A Pathway Analysis Experiment

Question

[Russh]

Hi were you reviewing a paper where ingenuity / metacore, or indeed HPRD/GO/etc, were used to identify coregulated clusters or functionally-related sets of genes from a microarray experiment what information would you expect to see explicitly stated in the (supplementary) methods?

To me it's not sufficient to say 'the identified genes were analysed using Ingenuity.'. Now, I'm being facetious, but I have seen this level of detail presented in papers (though typically outside of bioinfx journals).

I'd like to see a) date of access to the software/db; b) a clear definition of how genes were selected to be in the background set and 'active' sets on which the analysis was performed; c) any other user specified parameters.

However, I know that 'c)' above is rather ill-defined, so I was wondering whether any of you had any 'methods-section' recommendations for these kinds of experimetns.

Kind regards, R

Answer 1

[Chris Evelo]

In order to be able to redo the analysis I think you would have to have the following:

1. A clear description of the gene expressions statistics performed. Ideally the results of the gene statistics performed on the normalized data should be given. In principle with raw data, normalization method and statistics described you should be able to rebuild that though. This is just the input.
2. A clear description of the software version used and the pathway collection used. For WikiPathways that would for instance be "all pathways with the analysis collection curation tag downloaded on xxx".
3. Identification of what statistical results on the gene level were used as input for the statistics (p-values, fold changes or both in combination).
4. Identification of the statistical approach used (e.g. GSEA or just z-score calculations) and if these have settings (like cut-off for z-score), what were they?.
5. A clear description how the procedure applied deals (or does not deal) with the difference between upregulated and downregulated genes.
6. Most importantly keep in mind that actually pathway statistics are just a way to find the pathways that might be important! Biology affects the statistics a lot. I described that in an answer to another question. One example here. Adding non-regulated genes to a pathway will change the statistics, it does not change the biology... So, you will always want to see the really changed pathways. Just a list of affected pathways never is good enough.