Fit biofluid-residualized ALR — fit_biofluid_residualized

Fits one per-feature training model, `log(feature_count) ~ disease + biofluid + (1 | cohort)`, subtracts the estimated biofluid fixed effect at prediction time, and scores the panel as an ALR against a frozen pivot pool. If `lme4` is unavailable or a feature model fails, an `lm()` fallback is used and recorded.

Usage

fit_biofluid_residualized_alr(
  train_expr,
  train_meta,
  panel,
  biofluid_col = "biofluid",
  outcome_col = "disease",
  cohort_col = "cohort",
  pivot_features = NULL,
  pseudocount = NULL,
  min_pivot_present = 3L
)

Arguments

train_expr: Numeric matrix, training samples x features.
train_meta: Data frame aligned to `train_expr`.
panel: Character vector of panel features.
biofluid_col: Column identifying biofluid class.
outcome_col: Binary disease label column in `train_meta`.
cohort_col: Cohort column in `train_meta`.
pivot_features: Optional ALR pivot features.
pseudocount: Optional additive pseudocount before log transform.
min_pivot_present: Minimum pivot features required for prediction.