compare_groups: Compare numeric GRanges metadata across groups

In TmCalculator: Genome-Wide Nucleic Acid Melting Temperature Profiling and Multi-Omics Integration

Compare numeric GRanges metadata across groups

Description

Test whether one or more numeric metadata columns (e.g. Tm, GC) differ between levels of a categorical grouping column in a GRanges object.

Usage

compare_groups(
  gr,
  target = "Tm",
  method = c("wilcoxon", "t.test"),
  group,
  min_n_per_group = 2L,
  paired = FALSE,
  alternative = c("two.sided", "less", "greater"),
  posthoc = TRUE,
  p.adjust.method = c("holm", "hochberg", "hommel", "bonferroni", "BH", "BY", "fdr",
    "none")
)

Arguments

gr	A GRanges object.
target	Character vector of metadata column names to test (e.g. c("Tm", "GC")). All must be numeric.
method	Statistical test family. One of: "wilcoxon": rank-based tests (Wilcoxon / Kruskal-Wallis). "t.test": Gaussian-based tests (t-test / ANOVA).
group	Character. Name of a metadata column in gr defining groups.
min_n_per_group	Minimum number of non-missing observations required per group. Default: 2.
paired	Logical. Only for exactly two groups. Default: FALSE.
alternative	Character. Alternative hypothesis for two-group tests. One of "two.sided" (default), "less", or "greater".
posthoc	Logical. For \ge 3 groups, run pairwise follow-up tests with multiple-testing correction. Default: TRUE.
p.adjust.method	Multiple-testing correction for post-hoc pairwise comparisons. Passed to pairwise.wilcox.test / pairwise.t.test. One of: "holm" (default), "hochberg", "hommel", "bonferroni", "BH" (Benjamini-Hochberg FDR), "BY" (Benjamini-Yekutieli), "fdr" (alias of "BH"), or "none".

Details

The test used depends on method and the number of group levels:

Groups	method = "wilcoxon"	method = "t.test"
2	Wilcoxon rank-sum (or signed-rank if paired = TRUE)	Welch two-sample t-test (or paired t-test)
\ge 3	Kruskal-Wallis	One-way ANOVA (Welch)

When there are three or more groups and posthoc = TRUE, pairwise follow-up tests are run (Wilcoxon or Welch t-test) with p.adjust multiple-testing correction.

Value

A list with:

results

Data frame with one row per target: omnibus test name, statistic, p.value, and group information.

summary

Per-group descriptive statistics (n, mean, sd, median).

pairwise

Pairwise comparisons when posthoc = TRUE and there are \ge 3 groups; otherwise NULL.

Author(s)

Junhui Li

Examples

## Not run: 
library(GenomicRanges)
set.seed(1)
gr <- GRanges(
  seqnames = rep("chr1", 90),
  ranges = IRanges(start = sort(sample(1e6, 90)), width = 50),
  Tm = c(rnorm(30, 74, 2), rnorm(30, 70, 2), rnorm(30, 66, 2)),
  GC = runif(90, 40, 60)
)
gr$group <- rep(c("high", "mid", "low"), each = 30)

# Two groups
gr2 <- gr[gr$group %in% c("high", "low")]
compare_groups(gr2, target = "Tm", group = "group", posthoc = FALSE)

# Three or more groups (Kruskal-Wallis + pairwise Wilcoxon)
compare_groups(gr, target = c("Tm", "GC"), group = "group",
                     method = "wilcoxon")

# Three or more groups (one-way ANOVA + pairwise t-tests)
compare_groups(gr, target = "Tm", group = "group", method = "t.test")

# Post-hoc with Benjamini-Hochberg FDR control
compare_groups(gr, target = "Tm", group = "group",
                     p.adjust.method = "BH")

## End(Not run)

TmCalculator documentation built on Aug. 28, 2026, 5:09 p.m.