The five metrics

For two sets A and B of sizes |A|, |B| with intersection |A ∩ B| drawn from a universe of N items:

• Jaccard = |A ∩ B| / |A ∪ B|. Range [0, 1]. Symmetric.
• Dice = 2 |A ∩ B| / (|A| + |B|). Range [0, 1]. Symmetric. Always >= Jaccard; the two relate by Dice = 2J / (1 + J).
• Overlap coefficient = |A ∩ B| / min(|A|, |B|). Range [0, 1]. Equal to 1 when one set is contained in the other.
• Hypergeometric p-value — probability of observing |A ∩ B| or more shared items by chance, given |A|, |B|, and N. Tests over- representation.
• Fold enrichment = (|A ∩ B| * N) / (|A| * |B|). The ratio of observed to expected intersection size under independence. > 1 is over- representation.

Compute one metric directly

The helpers are exported and stateless:

jaccard(size_a = 138, size_b = 581, intersection = 100)
dice(size_a = 138, size_b = 581, intersection = 100)
overlap_coefficient(size_a = 138, size_b = 581, intersection = 100)
hypergeometric_p_value(N = 20000, K = 138, n = 581, k = 100)
fold_enrichment(N = 20000, K = 138, n = 581, k = 100)

The hypergeometric p-value is essentially zero: 100 shared genes out of an expected (138 * 581) / 20000 ≈ 4 is a 25× enrichment.

All pairs at once

statistics(result) returns five tables (four square NxN matrices for the ratio metrics + a long-form data.frame for the hypergeometric test):

stats@jaccard

head(stats@hypergeometric)

BH-FDR adjustment

stats@hypergeometric already carries the BH-FDR-adjusted q-value (p_adjusted) and a boolean significant (q < 0.05) and highly_significant (q < 0.001). The adjustment uses stats::p.adjust(method = "BH"):

raw_p <- stats@hypergeometric$p_value
adjusted <- bh_fdr(raw_p)
all.equal(adjusted, stats@hypergeometric$p_adjusted)

For unrelated p-values, BH-FDR is more permissive than Bonferroni and more conservative than no correction:

toy_p <- c(0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 0.9)
data.frame(
    raw            = toy_p,
    bonferroni     = pmin(toy_p * length(toy_p), 1),
    bh_fdr         = bh_fdr(toy_p)
)

broom-compatible long-form

broom::tidy() produces a tibble that’s pipeline-friendly (one row per pair, all metrics in one frame, sorted by adjusted p-value):

broom::tidy(result)

Reproducing the web tool’s coloring

The web tool colors significant pairs via the same p_adjusted thresholds:

sig_table <- broom::tidy(result)
sig_table$colour <- ifelse(sig_table$highly_significant, "red",
                    ifelse(sig_table$significant,        "orange",
                                                          "grey"))
sig_table[, c("set_a", "set_b", "p_adjusted", "colour")]

What’s next

• vignette("v02_real_cancer_drivers") — see these stats in the context of a real biological analysis.
• vignette("v06_pipeline_integration") — feed broom::tidy() into a downstream tidyverse pipeline.