Build precision-weight input from a curveR calibration result

Converts a curveRcore calibration_result or calibration_result_multiplate (the object returned by curveRfreq::fit_calibration_freq_multiplate() or curveRbayes::fit_calibration_bayes()) into the tidy, standardized data frame consumed by fit_precision_weights(). A plain data frame is also accepted (validation/passthrough), which is the escape hatch for the original foreign-data workflow.

Usage

as_weight_data(
  x,
  design,
  source = c("samples", "grid"),
  conc_scale = c("log10", "natural"),
  include_plate = TRUE,
  plate_in_cell = FALSE,
  drop_oor = FALSE,
  ...
)

# S3 method for class 'calibration_result_multiplate'
as_weight_data(
  x,
  design,
  source = c("samples", "grid"),
  conc_scale = c("log10", "natural"),
  include_plate = TRUE,
  plate_in_cell = FALSE,
  drop_oor = FALSE,
  ...
)

# S3 method for class 'calibration_result'
as_weight_data(
  x,
  design,
  source = c("samples", "grid"),
  conc_scale = c("log10", "natural"),
  include_plate = TRUE,
  plate_in_cell = FALSE,
  drop_oor = FALSE,
  ...
)

# S3 method for class 'data.frame'
as_weight_data(
  x,
  design,
  source = c("samples", "grid"),
  conc_scale = c("log10", "natural"),
  include_plate = TRUE,
  plate_in_cell = FALSE,
  drop_oor = FALSE,
  ...
)

Arguments

x

A calibration_result, calibration_result_multiplate, or data.frame.

design

Character vector of design-group column names that define the saturated cells (e.g. c("timeperiod", "cohort_arm")). Required, and must be present on the input, when source = "samples": for a calibration_result(_multiplate) these are carried through from the original samples data frame passed at fit time — if they are absent, supply them on the fitting input first, or use the data.frame method with a pre-joined frame. Ignored (silently intersected with the available columns) when source = "grid": see source below.

source

"samples" (default) extracts per-sample predictions, validates that every design column is present, and builds the .cell saturated-cell factor used by fit_precision_weights(). "grid" extracts the precision grid (used by predict_weights() to build a continuous weight profile): curveRcore::tidy_grid() returns a per-curve concentration profile, not a per-design-cell table, so it never carries the original samples design columns. design and the .cell/within-cell-replication checks therefore do not apply to source = "grid" — the profile only needs se/concentration/pcov, which predict_weights() consumes directly.

conc_scale

Location-response scale: "log10" (default, uses predicted_concentration) or "natural" (uses final_concentration when present, else 10^predicted_concentration).

include_plate

Logical: attach curve_id as a column and make it available as a plate grouping. Default TRUE.

plate_in_cell

Logical: fold curve_id into the cell factor alongside design. Default FALSE — plate variation is left to be absorbed by phi / the plate random effect (see the plan, §5).

drop_oor

Logical: drop rows with pcov_pass == FALSE? Default FALSE (keep but warn) — not discarding out-of-range observations is the point of the package.

...

Passed to methods.

Value

A data frame (class weight_data) with at least: obs_id, sampleid (if present), curve_id (if include_plate), concentration (location response on conc_scale), predicted_concentration (log10, for the estimator), se (= se_concentration, the canonical uncapped scale predictor), pcov (reference only), pcov_pass, and, when source = "samples" (or any design columns are present on a "grid" input), the design columns and .cell. Carries attributes conc_scale, is_log_independent, design (resolved to the columns actually present), plate_in_cell.

Details

Extraction from the S3 object is delegated to curveRcore::tidy_samples() / curveRcore::tidy_grid() so that the data contract has a single owner; this adapter never reaches into object internals.

See also

fit_precision_weights(), predict_weights(), curveRcore::tidy_samples()