Skip to contents

Get calibration metrics

Source: R/calc-calibrations.R
get_calibration_metrics.Rd

Extracts calibration fit metrics from a MRMhubExperiment object.

Usage

get_calibration_metrics(
  data = NULL,
  include_qualifier = TRUE,
  with_lod = TRUE,
  with_loq = TRUE,
  with_coefficients = TRUE,
  with_sigma = TRUE,
  summary_table = FALSE
)

Arguments

data

A MRMhubExperiment object with QC metrics.

include_qualifier

Whether to include qualifier features. When FALSE, only quantifier features are returned. Default is TRUE.

with_lod

Whether to include LoD in output. Default is TRUE.

with_loq

Whether to include LoQ in output. Default is TRUE.

with_coefficients

Whether to include regression coefficients. Default is TRUE.

with_sigma

Whether to include sigma in output. Default is TRUE.

summary_table

When TRUE, return a compact, display-formatted summary tibble (analyte, fit_model, fit_weighting, r2, lod, loq) with r2 rounded to 5 decimals and lod/loq to 3 significant figures. This is a presentation view for reporting; the values are rounded, so do not use it for downstream computation. lod/loq follow with_lod/with_loq. The with_coefficients/with_sigma flags are ignored in this mode. Default is FALSE (full, unrounded metrics).

Value

A tibble with exported calibration metrics.

Details

Requires prior computation of regression results using calc_calibration_results(). See its documentation for details.

Returned Details and Metrics

  • feature_id: Feature identifier.

  • is_quantifier: Logical, indicates if the feature is a quantifier.

  • fit_model: Regression model used for fitting.

  • fit_weighting: Weighting method used in fitting.

  • lowest_cal: Lowest nonzero calibration concentration.

  • highest_cal: Highest calibration concentration.

  • r2: R² value, indicating goodness of fit. For a weighted fit this is the weighted coefficient of determination (computed from weighted sums of squares), matching the value reported by vendor software such as Agilent MassHunter for the same weighted curve.

  • coef_a: Intercept (0th-order term, x^0) of the fitted curve.

  • coef_b: 1st-order (linear, x^1) coefficient — the slope in a linear fit, or the linear term in a quadratic fit.

  • coef_c: 2nd-order (quadratic, x^2) coefficient in quadratic fits. Returns NA for linear fits.

  • sigma: Standard deviation of residuals.

  • reg_failed: TRUE if regression fitting failed.

  • LoD = 3.3× the sample standard error of residuals / slope of the regression (see Notes).

  • LoQ = 10× the sample standard error of residuals / slope of the regression (see Notes).

The coefficients describe the curve in ascending power order (R's lm() / poly() convention): response = coef_a + coef_b * x + coef_c * x^2. This is the reverse of the descending a*x^2 + b*x + c form used by some vendor software (e.g. Agilent MassHunter, where a is the x^2 term), so when comparing against such an export, match by power, not by letter.

Note: LoD/LoQ follow the ICH Q2(R1/R2) approach (3.3 sigma / S and 10 sigma / S). The slope S is the slope of the calibration curve at zero concentration (the linear coefficient coef_b); for a quadratic fit the quadratic term does not contribute to this slope. The response sigma is selectable in calc_calibration_results() via lod_sigma (residual standard error, the default, or the standard error of the intercept); the sigma column reported here is always the residual standard error.

For a weighted fit (1/x, 1/x^2, 1/sqrt(x)) sigma is R's weighted residual standard error, which is not on the raw response scale that the ICH 3.3 sigma / S formula assumes, so the reported LoD/LoQ are approximate (typically slightly optimistic for 1/x). Use fit_weighting = "none" if you require the strict ICH response-scale Sy/x; the back-calculated concentrations themselves are unaffected.