Fit disease weights from a cohort/provenance mixed-effects model — fit_mixed_effects

Fits one per-feature model on the training pool: `log(feature_count) ~ disease + (1|cohort) + (1|provenance_block)`. Terms with only one level are omitted. If `lme4` cannot fit a feature, a fixed-effect `lm()` fallback is attempted and recorded in the fit audit.

Usage

fit_mixed_effects_scorer(
  train_expr,
  train_meta,
  panel,
  outcome_col = "disease",
  cohort_col = "cohort",
  block_col = "provenance_block",
  pseudocount = NULL
)

Arguments

train_expr: Numeric matrix, training samples x features.
train_meta: Data frame aligned to `train_expr`.
panel: Character vector of panel features.
outcome_col: Binary disease label column in `train_meta`.
cohort_col: Training cohort column.
block_col: Training provenance-block column.
pseudocount: Optional additive pseudocount before log transform.

Value

A fit object consumed by `predict_mixed_effects_scorer()`.