Skip to contents

Select differential edges with Bayesian FDR control

select_diff_edges.Rd

Computes a per-edge posterior probability that two groups differ and selects edges with bayes_fdr(). Chains are pooled by default.

Usage

select_diff_edges(
  fit,
  groups = c(1, 2),
  alpha = 0.05,
  rule = c("magnitude", "structural"),
  delta = 0.05,
  scale = c("pcor", "precision"),
  chain = NULL
)

Arguments

fit

A multiggm_fit or multiggm_fit_list object.

groups

Integer vector of length 2, the groups c(k, m) to compare. Default c(1, 2).

alpha

Target posterior expected FDR. Default 0.05.

rule
"magnitude"

(default) \(P(|D_{ij}| > \delta)\), where \(D\) is the group difference on scale. Works for every sampler. FDR is controlled for the hypothesis \(|D_{ij}| > \delta\): a true difference smaller than delta counts as a null.

"structural"

\(P(g_{k,ij} \ne g_{m,ij})\), the posterior probability that the edge is in exactly one of the two graphs. Needs edge-inclusion indicators, so it is not available for mGHS fits.

delta

Magnitude threshold for rule = "magnitude", on the scale given by scale. Default 0.05.

scale

"pcor" (default) for partial correlations or "precision" for precision-matrix entries.

chain

NULL (default) pools all chains; an integer vector selects chains of a multiggm_fit_list.

Value

An object of class "multiggm_diff_selection": a list with selected (symmetric logical [p, p] matrix), prob (symmetric [p, p] matrix of the edge probabilities used), edges (data frame of the selected edges with i, j, prob and, for the magnitude rule, mean difference), expected_fdr, n_selected, alpha, rule, delta, scale and groups.

See also

diff_summary(), bayes_fdr(), pip_diff_edge()

Examples

sim <- simulate_multiggm(K = 2, p = 8, n = 80, seed = 1)
fit <- multiggm_mcmc(data_list = sim$data_list, burnin = 200, nsave = 100)
sel <- select_diff_edges(fit, alpha = 0.1, delta = 0.1)
sel
#> <multiggm_diff_selection>
#>   Groups 1 vs 2, rule: magnitude (P(|D| > 0.1), pcor scale)
#>   Selected 7 edges at target FDR 0.1 (posterior expected FDR 0.083)
#>   i j prob        mean
#> 1 1 2 1.00 -0.54583861
#> 2 2 4 1.00  0.41929256
#> 3 2 5 1.00  0.48157263
#> 4 3 4 0.98  0.49929607
#> 5 7 8 0.93  0.32794996
#> 6 4 6 0.89 -0.24818914
#> 7 6 7 0.62  0.06962387