Fit kit-residual MAD scoring. — fit_kit_residual

Per feature, log-CPM values are robustly centered and scaled within kit/biofluid strata using median and MAD. Unknown kit falls back to biofluid, then global. This optional method is included as a negative or robustness comparator.

Usage

fit_kit_residual_mad(
  expr_matrix,
  sample_meta,
  kit_label_col,
  biofluid_col = "biofluid",
  min_samples_per_stratum = 5L,
  pseudocount = 0.5,
  mad_floor = 0.001
)

Arguments

expr_matrix: Numeric matrix, samples x features, counts or non-negative abundance values.
sample_meta: data.frame with one row per sample.
kit_label_col: Column in `sample_meta` containing library-kit family. Missing/unknown kit values fall back to `biofluid_col`, then global.
biofluid_col: Optional fallback column. Default "biofluid".
min_samples_per_stratum: Minimum training samples needed for a stratum-specific centering set.
pseudocount: Additive pseudocount before log-CPM transformation.
mad_floor: Minimum robust scale used to avoid division by zero.

Value

Fit object consumed by `score_kit_residual_mad()`.