curveR

Calibration curve ecosystem for quantitative immunoassays.

What is curveR?

curveR is a suite of four R packages that share a common data contract, identical model definitions, and a common output class for immunoassay calibration curve analysis. Frequentist and Bayesian results are directly comparable — fitting engine is a single function-call difference.

The four packages

Package	What it does	Install if you want
curveRcore	Shared models, inverses, derivatives, preprocessing, output class	Always required
curveRfreq	Frequentist multi-start Levenberg–Marquardt NLS calibration	Fast fitting, no Stan
curveRbayes	Bayesian hierarchical calibration via Stan (4PL/5PL/Gompertz)	Full posterior uncertainty, multi-plate pooling
curveRweights	Continuous precision weights from calibration-curve uncertainty	Weighted downstream analysis instead of a binary LLOQ/ULOQ gate

Installation

# Full ecosystem — installs curveRcore + curveRfreq immediately;
# curveRbayes and curveRweights available once CmdStan is installed.
remotes::install_github("immunoplex/curveR")

# Stan backend (required for curveRbayes and curveRweights)
install.packages("cmdstanr",
                 repos = c("https://stan-dev.r-universe.dev",
                           getOption("repos")))
cmdstanr::install_cmdstan()

# Frequentist only — no Stan needed
remotes::install_github("immunoplex/curveRcore")
remotes::install_github("immunoplex/curveRfreq")

# Add precision weighting after the above
remotes::install_github("immunoplex/curveRweights")

Quick start

library(curveR)   # attaches curveRcore + curveRfreq
library(curveRcore)

data(bead_assay_example, package = "curveRcore")

# Preprocess standards
std_pre <- curveRcore::preprocess_standards(
  data                 = bead_assay_example$standards,
  response_variable    = "mfi",
  independent_variable = "dilution",
  is_log_response      = TRUE,
  is_log_independent   = TRUE,
  std_curve_conc       = 10000
)

# Fit frequentist curves for all curve_ids
mp_freq <- curveRfreq::fit_calibration_freq_multiplate(
  standards          = std_pre$data,
  samples            = bead_assay_example$samples,
  response_var       = "mfi",
  is_log_response    = TRUE,
  is_log_independent = TRUE,
  std_curve_conc     = 10000
)

# One-row-per-curve summary
curveRfreq::summary_table(mp_freq)

# Fit Bayesian curves (requires CmdStan)
if (requireNamespace("curveRbayes", quietly = TRUE)) {
  mp_bayes <- curveRbayes::fit_calibration_bayes(
    standards          = std_pre$data,
    samples            = bead_assay_example$samples,
    response_var       = "mfi",
    is_log_response    = TRUE,
    is_log_independent = TRUE,
    std_curve_conc     = 10000,
    seed               = 42
  )
  # Compare frequentist vs Bayesian back-calculations
  curveRcore::compare_calibrations(mp_freq, mp_bayes)
}

# Precision weighting — samples must carry design columns
if (requireNamespace("curveRweights", quietly = TRUE)) {
  wd <- curveRweights::as_weight_data(
    mp_freq, design = c("timeperiod", "cohort_arm")
  )
  pw <- curveRweights::fit_precision_weights(wd)
  pw
}

---

Methodology vignette

The full manuscript-quality methods comparison — preprocessing, model forms, NLS engine, hierarchical Stan model, MCMC diagnostics, LOO-CV selection, CDAN precision profiling, sample back-calculation, agreement assessment, and precision-weighted downstream analysis — is in:

vignette("curveR-methods-comparison", package = "curveR")

This vignette lives exclusively in the hub because it requires all four packages simultaneously.

---

CRAN and r-universe availability

Package	CRAN	r-universe / GitHub
curveRcore	Planned	✔ github.com/immunoplex/curveRcore
curveRfreq	Planned	✔ github.com/immunoplex/curveRfreq
curveRbayes	r-universe + GitHub only (CmdStan dependency)	✔ github.com/immunoplex/curveRbayes
curveRweights	r-universe + GitHub only (CmdStan dependency)	✔ github.com/immunoplex/curveRweights

License

AGPL-3.0