Per-marker distributional drift between acquisition batches

For each marker, the largest Earth Mover's distance between any two batches, reported in analysis units and scaled by the marker's pooled MAD so markers can be compared with each other.

Usage

marker_batch_drift(
  cells,
  batch_col,
  markers = NULL,
  max_cells = 20000L,
  min_cells = 200L,
  flag_at = 0.5,
  seed = 42L
)

Arguments

cells: cell-level table carrying sample_id, the batch column, and one column per marker on the analysis scale
batch_col: name of the batch column in cells
markers: marker columns to test; defaults to every numeric column that is not structural
max_cells: cells sampled per batch before the distances are computed; the quantile function stops moving long before the whole batch is used
min_cells: smallest batch worth comparing
flag_at: value of emd_over_mad at or above which a marker is flagged
seed: seed for the local RNG stream, which is restored on exit

Value

a data.frame ordered by emd_over_mad, or NULL

Details

WHAT A FLAG MEANS. A marker whose distribution differs between batches by an appreciable fraction of its own spread was measured differently in them. That is a statement about the assay, not about the donors, unless batch and study group coincide – which is the question batch_mixing_report() answers with Cramer's V, and which should be read first. Where the two overlap, this table cannot separate a reagent lot from the biology either.