Spreading across a cohort, and which markers it is costing

Aggregates spreading_pairs() over samples and joins the result to how each receiver's threshold was obtained. A marker that receives a great deal of spreading AND falls back to a quantile is the actionable finding: its cut is unresolved for a reason the panel can fix.

Usage

run_spreading_report(
  pops,
  gates,
  thr_all = NULL,
  panel_of = NULL,
  max_samples = 8L,
  flag_ratio = 1.25,
  trend_flag = 0.3,
  exclude = character(0)
)

Arguments

pops

per-sample scoring results, carrying tmat and thresholds

gates

per-sample gate objects

thr_all

the thresholds table, used for the fallback rate per marker

panel_of

named character vector mapping sample to panel

max_samples

cap on samples contributing, since the ranking stabilises long before every sample is used

flag_ratio

spreading ratio at or above which a pair is called substantial

trend_flag

absolute rank correlation at or above which a pair is flagged as a possible compensation or unmixing error. Default 0.3, a conventional moderate correlation, not a validated cut. A negative trend on a pair whose receiver is flag_ratio times wider at a given source level is reported as trend_unresolved instead, since truncation at the receiver's cut produces one from spreading alone.

exclude

sample ids kept out: declared controls and staining-QC failures, which are not samples of the cohort.

Value

list(pairs, receivers), each a data.frame, or NULL