Statistics

Question

[erinda]

Dear all.

I would like to ask a question, and would appreciate your help.

I performed a pangenome analysis using Roary on bacterial isolates from diseased and healthy cohorts, followed by association analysis with Scoary. For a specific gene, the scoary results showed:

Prevalence: 31.3% in diseased cohort vs. 11% in healthy individuals
Odds ratio: 3.68 (95% CI: 1.36–9.96)
Fisher's exact test: p = 0.012
Empirical p-value (10,000 permutations): p = 0.0039
Adjusted p-values: Benjamini–Hochberg p = 1.0; Bonferroni p = 1.0
Pairwise comparisons: Best p = 0.18; Worst p = 0.79

The Fisher's exact test, empirical permutation test, and 95% CI all suggest significance, but standard multiple testing corrections (Bonferroni, Benjamini–Hochberg) are non-significant.

How should I interpret these results? Is this gene associated with disease, or should this be considered only a suggestive trend?

Any help in interpreting these results ?

Thank you very much in advance!

Answer 1

[andres.firrincieli]

Not associated with disease because it did not pass the multiple testing correction. What you’re seeing is likely a false positive.

edit: see ATpoint answer

[ATpoint]

I think "no significance for association" should be the statement, because the test does not assess significance for absence of association. It tests for association, but there is no significance for that. Multiple testing adjusted pvalues are a bit hard to interpret because at some point all turn 1 and that depends both on the actual pvalue and all other pvalues in the analysis.