Fit Bayesian Hierarchical Calibration Curves — fit_calibration

Fits a hierarchical Bayesian model to preprocessed standard curve data across one or more curve_ids simultaneously. Returns a calibration_result_multiplate with one entry per curve_id, each containing its own grid predictions, sample predictions, and parameter summaries.

Usage

fit_calibration_bayes(
  standards,
  samples = NULL,
  blanks = NULL,
  response_var,
  model_names = c("logistic4", "gompertz4"),
  is_log_response = TRUE,
  is_log_independent = TRUE,
  std_curve_conc,
  fixed_a = NULL,
  cv_x_max = 150,
  pcov_threshold = 20,
  min_dynamic_range_log10 = 0.5,
  max_rel_se = 5,
  n_grid = 200L,
  grid_min_conc = 1e-04,
  grid_max_conc = NULL,
  chains = 4L,
  warmup = 1000L,
  sampling = 1000L,
  adapt_delta = 0.9,
  seed = NULL,
  n_draws_predict = 500L,
  n_draws_ensemble = 260L,
  compute_all_grids = FALSE,
  use_heteroscedastic_noise = FALSE,
  run_loo = NULL,
  verbose = FALSE
)

Arguments

standards: Data frame. Preprocessed stacked standard curve data. Must contain curve_id, a response column, and a concentration column — all on the fitting scale.
samples: Data frame or NULL. Stacked sample data with curve_id and the response column (on the raw measurement scale).
blanks: Data frame or NULL. Blank well data with curve_id and the response column (on the fitting scale). When supplied, blanks are passed to Stan to anchor the lower asymptote via a separate likelihood term, and are stored in each per-curve calibration_result$blanks slot for downstream QA. Default NULL.
response_var: Character. Name of the response column.
model_names: Character vector. Models to fit. Default c("logistic4", "gompertz4").
is_log_response: Logical. Default TRUE.
is_log_independent: Logical. Default TRUE.
std_curve_conc: Numeric. Undiluted standard concentration.
fixed_a: Numeric or NULL. Fixed lower asymptote (fitting scale).
cv_x_max: Numeric. Default 150.
pcov_threshold: Numeric. Percent CV threshold for LLOQ/ULOQ determination and the dynamic-range eligibility gate. Default 20.
min_dynamic_range_log10: Numeric. Minimum dynamic range (log10) for eligibility. Default 0.5.
max_rel_se: Numeric. Maximum relative SE (SD/|mean|) permitted for any parameter. Default 5.0.
n_grid: Integer. Default 200.
grid_min_conc: Numeric. Default 1e-4.
grid_max_conc: Numeric or NULL.
chains: Integer. Default 4.
warmup: Integer. Default 1000.
sampling: Integer. Default 1000.
adapt_delta: Numeric. Default 0.9.
seed: Integer or NULL.
n_draws_predict: Integer. Number of posterior draws for the best-model grid and sample predictions. Default 500.
n_draws_ensemble: Integer. Number of posterior draws for non-best-model precision grids. Default 260.
compute_all_grids: Logical. If TRUE, compute precision grids for every converged model. Required for eligibility gating when more than one model is fitted. Default FALSE.
use_heteroscedastic_noise: Logical. If TRUE, the Stan models use a power-of-mean variance function sigma_i = exp(log_sigma0 + log_sigma_slope * log(|mu_i|)) in the likelihood, and the same sigma_i is injected when generating the CDAN noisy observations in predict_grid_bayes(). This restores the O'Malley (2008) CDAN precision profile interpretation. If FALSE (default), a constant sigma_obs is used and the precision profiles reflect posterior-predictive uncertainty driven mainly by inverse- curve geometry.
run_loo: Logical or NULL. Default NULL (auto).
verbose: Logical. Default FALSE.

Value

A calibration_result_multiplate object (from curveRcore). Each per-plate $selection contains $assessments, $eligible_models, and $fallback from the eligibility gating. Each per-plate calibration_result carries $standards and $blanks slots with the per-curve subsets of the input data.