Fit technology-residualized ALR — fit_tech_residualized

Per feature, the training model is `log(feature_count) ~ disease + technology + (1|cohort)` when `lme4` is available and more than one cohort exists. Prediction subtracts the learned technology fixed effect and then computes an ALR-style panel score.

Usage

fit_tech_residualized_alr(
  train_expr,
  train_meta,
  panel,
  technology_col = "technology",
  outcome_col = "disease",
  cohort_col = "cohort",
  pivot_features = .ta_default_pivot_pool(),
  anchor_features = character(),
  min_pivot_present = 3L,
  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.
technology_col: Column identifying technology class.
outcome_col: Binary disease label column in `train_meta`.
cohort_col: Cohort column in `train_meta`.
pivot_features: Candidate ALR pivot features.
anchor_features: Optional train-selected fallback pivots.
min_pivot_present: Minimum pivots required for prediction.
pseudocount: Optional additive pseudocount before log transform.

Value

Fit object consumed by `predict_tech_residualized_alr()`.