Summarize systematic prevalence evaluation results and recommend a threshold

Usage

summarize.prev.eval(
  results,
  unstable.delta.nsig = 8L,
  unstable.frac.nsig = 0.5,
  stable.window = 2L,
  stable.max.delta.nsig = 4L,
  b.min.features = 30L,
  b.min.sig = 5L,
  c.target.features = 35L,
  c.min.sig = 5L,
  avoid.unstable = TRUE
)

Arguments

results: A results object returned by basin.prev.eval(). Must contain a data.frame results$summary with columns: prevalence, nFeatures, nSig, and (optionally) medianAbsLogFC.
unstable.delta.nsig: Integer; flag an "unstable" step if the absolute change in nSig between adjacent prevalence thresholds is >= this value.
unstable.frac.nsig: Numeric in $[0,1]$; alternative instability criterion: flag if abs(delta.nSig) >= unstable.frac.nsig * max(nSig.prev, nSig.next). Either criterion triggers a flag.
stable.window: Integer; window size (number of adjacent steps) used by Rule A (stability-first) to check local stability.
stable.max.delta.nsig: Integer; for Rule A, require that abs(delta.nSig) is <= this value across the next stable.window steps.
b.min.features: Integer; for Rule B, only consider thresholds with nFeatures >= this value.
b.min.sig: Integer; for Rule B, only consider thresholds with nSig >= this value.
c.target.features: Integer; for Rule C, aim for approximately this many retained features (interpretability-first).
c.min.sig: Integer; for Rule C, require at least this many discoveries.
avoid.unstable: Logical; if TRUE, avoid selecting thresholds that are adjacent to unstable jumps.

Value

A list with:

summary.aug: results$summary augmented with sig.rate and deltas
unstable.steps: integer indices of steps flagged as unstable (between i-1 and i)
unstable.prevalence: prevalence values at the right end of unstable steps
recommend: list with rule.a, rule.b, rule.c recommended prevalence values
recommend.idx: corresponding row indices in summary.aug