[WebGestalt WebServer] Discrepances in results in ORA, when a single vs multiple enrichment categories are selected

[Vasileios Zogopoulos (Bill)]

Dear WebGestalt Team,

We have been using WebGestalt extensively for enrichment analyses, as we find it to be the most detailed and consistent tool available.

However, we have noticed a discrepancy in the ORA results on the web server. Specifically, the "Expect", "Ratio", "P Value" and "FDR" values differ when selecting a single category versus selecting multiple categories simultaneously, when using the same gene list, background, and default settings (with a BH < 0.05 cutoff).

For example, using the "Gene/GO:BP" example dataset (500 genes):

1. Single Run: We run ORA selecting only  "geneontology - GO:Biological Process".

2. Combined Run: We run ORA selecting multiple categories by clicking the "+" button ("geneontology - GO:Biological Process" and adding "geneontology - GO:CC", "pathways – Reactome" and "network - Transcription Factor target" ).

When we compare the specific "GO:BP" terms from the combined run against the single run, the aforementioned metrics differ.

We were wondering if this is expected behavior.

Attached are screenshots of the single-category and multiple-category results for comparison.

Thank you in advance for your response.

Best Regards,

Vasileios

[John Elizarraras]

Hello Vasileios,

Thank you for reaching out.

This is intended behavior, albeit potentially confusing. Before running ORA enrichment, both the reference list and interesting list are filtered to only contain genes present in the gene set database. This is necessary to prevent poorly annotated genes having an unfair effect on the final statistics. When you add more gene sets to the analysis, the list of accepted genes increases, which is why the analysis is different between the single and multicategory runs. So, in the end, the reference and interesting lists are not the same between the single and multicategory runs.

 In general, this difference won't have a large effect on the result, but if you started with a small gene set list and then add a larger database, this could change your results more drastically.

Let me know if anything is unclear or if you have any more questions.

Best,

John