Batch-mixing diagnostic with a permutation null and a confounding check

Batch-mixing diagnostic with a permutation null and a confounding check

Usage

batch_mixing_report(
  cells,
  batch_col,
  group_col = "cohort",
  k = 30L,
  max_cells = 4000L,
  n_perm = 20L,
  seed = 42L
)

Arguments

cells

embedding cell table with umap_1/umap_2

batch_col

column naming the batch (acquisition date, run, operator)

group_col

one or more columns to test batch against; one row of the confounding table per column. Pass the study group AND the timepoint on a repeated-measures design: the visit is the primary comparison there, and checking only the study group leaves it unexamined.

k

neighbourhood size

max_cells

subsample ceiling; LISI is a local statistic and converges quickly, so a few thousand cells is ample and keeps this from dominating the runtime of the whole pipeline

n_perm

permutations for the null

seed

Random seed. The RNG stream is restored afterwards, so this cannot perturb later sampling. Default 42L.

Value

list(summary = data.frame, per_cell = numeric, confounding = data.frame)