Simplify Fisher's test section in GO over-enrichment analysis episode

[js2264]

Regarding the Fisher's exact test section, I would propose to largely simplify the section by presenting the problem through 2 questions : "1) for every gene in my universe, is the gene DE (TRUE/FALSE)? ; and 2) for every gene in my universe, is the gene in my gene set of interest (TRUE/FALSE)?". This can be done by creating a 2-column data.frame, then doing a table of it.

universe = rownames(se)
df <- data.frame(
    isSE = universe %in% sexDEgenes, 
    isGS = universe %in% XYGeneSet
)
# then 
table(df) |> fisher.test()

This has several advantages:

• Explicitly requiring to define the universe (currently, the n value is used to calculate the intermediate n_20 and n_02 values, but not explicitly when filling out the matrix for the test).
• Using table to do the calculation, rather than manually trying to calculate each value by hand (1 operation, vs 9 individual operations with not-so-easy variable names) -> less error-prone

This has the downside that it's not as easy to explain how to calculate the calculation for phypher section (or maybe actually easier? I don't know):

# Question: are DE genes over-represented within a gene set of interest? 
1 - phyper(
    sum(universe %in% sexDEgenes & rownames(se) %in% XYGeneSet) - 1,           # genes DE AND in gene set of interest
    sum(universe %in% sexDEgenes),       # genes DE
    sum(! universe %in% sexDEgenes),     # genes not DE
    length(XYGeneSet)                                # gene set of interest
)

Let me know if you would like this to be changed, in which case I can work on a PR.

[csoneson]

I agree that table() seems easier (and the schematic tables can still be kept to match the notation with the observed numbers)

[jokergoo]

The use of n_* notations is mainly for explaining the Fisher's exact test and hypergeometric distribution, not for use in practice. From the perspective of teaching, maybe table(df) |> fisher.test() is not easy to explain to the students.