Score a panel after cohort-wise robust z-standardization — score_cohort_z_then_pool

Within each cohort, all panel features are log-transformed, centered by the cohort median, scaled by cohort MAD, and summed. When `fit` lacks a new test cohort, the default is to compute unsupervised moments from that test cohort; no disease labels are used for this fallback.

Usage

score_cohort_z_then_pool_score(
  expr,
  meta,
  panel,
  cohort_col = "cohort",
  block_col = "provenance_block",
  fit = NULL,
  allow_new_group_moments = TRUE,
  pseudocount = NULL
)

Arguments

expr: Numeric matrix, samples x features.
meta: Data frame aligned to `expr`.
panel: Character vector of panel features.
cohort_col: Column in `meta` identifying cohorts.
block_col: Column in `meta` identifying provenance blocks; accepted for a common scorer signature but not used by this method.
fit: Optional object from `fit_cohort_z_then_pool_score()`.
allow_new_group_moments: Logical; if TRUE, new cohorts are centered on their own unlabeled samples.
pseudocount: Optional additive pseudocount before log transform.

Value

Numeric score vector, one value per sample.