Score every subject against class-specific reference banks

Builds the subject-by-class matrix of class-conditional log-likelihoods used by the MBMM gate model.

Usage

mbmm_score_subjects(
  obs_list,
  bank_pred_list,
  sigma,
  donor_col = "id",
  bundle_col = "bundle_id",
  log_scale = TRUE,
  donor_weighting = c("equal", "particle"),
  loo = TRUE,
  verbose = TRUE
)

Arguments

obs_list

Named list of subject observation data.frames.

bank_pred_list

Named list of reference-bank prediction tables, one per bank-based class.

sigma

Measurement-error SD passed to mbmm_class_loglik.

donor_col

Name of the donor column in each bank table.

bundle_col

Name of the bundle column in each bank table.

log_scale

Whether observations, predictions, and LODs are already on the log10 likelihood scale.

donor_weighting

Either "equal" or "particle".

loo

Logical scalar or named logical vector. If a scalar, the same LOSO rule is applied to every class. If named, names must correspond to bank_pred_list.

verbose

Print progress every ten subjects.

Value

A data.frame with one row per subject and one logL_<class> column per bank.

Details

Leave-one-subject/donor-out scoring is controlled by loo; it is not applied automatically unless requested. This allows the same function to reproduce application-specific choices:

• SDY2354: typically loo = TRUE;

• COVID5001 productive bank: use the setting specified by the final manuscript analysis.