curveRbayes • curveRbayes

Bayesian hierarchical calibration curves for the curveR suite.

Installation

# Install curveRcore first
devtools::install_github("immunoplex/curveRcore")
# Then curveRbayes
devtools::install_github("immunoplex/curveRbayes")

Stan toolchain

curveRbayes fits models via cmdstanr. Install it once after installing the package:

install.packages("cmdstanr",
                 repos = c("https://stan-dev.r-universe.dev",
                           getOption("repos")))
cmdstanr::install_cmdstan()   # downloads and builds CmdStan

Quick start

library(curveRbayes)

data(bead_assay_example, package = "curveRbayes")

# Preprocess standards (upstream of fitting)
std_pre <- curveRcore::preprocess_standards(
  data                 = bead_assay_example$standards,
  antigen_settings     = antigen_settings,
  response_variable    = bead_assay_example$response_var,
  independent_variable = bead_assay_example$indep_var,
  is_log_response      = TRUE,
  is_log_independent   = TRUE
)

# Fit all curves simultaneously
mp <- fit_calibration_bayes(
  standards          = std_pre$data,
  samples            = bead_assay_example$samples,
  response_var       = "mfi",
  model_names        = c("logistic4", "gompertz4"),
  std_curve_conc     = 30,
  chains             = 4L,
  warmup             = 1000L,
  sampling           = 1000L,
  seed               = 42
)

# One-row-per-curve summary
summary_table_bayes(mp)

# All sample back-calculations
collect_samples_bayes(mp)

Quick Reference

Task	Command
Regenerate man pages	`devtools::document()`
Full site rebuild	`pkgdown::build_site(dest_dir = "docs")`
Reference pages only	`pkgdown::build_reference()`
Vignette only	`pkgdown::build_articles()`
Home page only	`pkgdown::build_home()`
Check no topics are unassigned	`setdiff(pkg$topics$name, unlist(lapply(cfg$reference, "[[", "contents")))`
Preview locally	Open `docs/index.html`
Push and deploy	`git add docs/ && git commit -m "..." && git push`
Check CmdStan installation	`cmdstanr::cmdstan_version()`
Recompile a Stan model	`compile_stan_model("logistic4")`

Stan models

curveRbayes ships five pre-written Stan models in inst/stan/. All use a non-centred parameterisation (NCP) to eliminate Neal’s funnel geometry and a reduce_sum map-reduce likelihood for multi-core speedup.

File	Model family	Parameters
`hierarchical_logistic4.stan`	4-parameter logistic	A, B, C, D
`hierarchical_logistic5.stan`	5-parameter logistic	A, B, C, D, G
`hierarchical_loglogistic4.stan`	4-parameter log-logistic	A, B, C, D
`hierarchical_loglogistic5.stan`	5-parameter log-logistic	A, B, C, D, G
`hierarchical_gompertz4.stan`	4-parameter Gompertz	A, B, C, D

When is_log_independent = TRUE and is_log_response = TRUE, loglogistic4 is mathematically equivalent to logistic4 and is automatically dropped from the candidate set.

See vignette("bayesian-quickstart", package = "curveRbayes") for the full worked example including Stan model internals, MCMC diagnostics, LOO-CV model selection, and the CDAN precision profiling procedure.

Troubleshooting

Problem	Fix
`curveRcore` not found during CI	Add `install_github("immunoplex/curveRcore")` before `setup-r-dependencies` in the workflow
`cmdstanr` not found	Install from Stan universe: `install.packages("cmdstanr", repos = c("https://stan-dev.r-universe.dev", getOption("repos")))`
CmdStan not installed	Run `cmdstanr::install_cmdstan()`
Stan file not found (`mustWork = TRUE` error)	Run `devtools::install()` so `inst/stan/` is copied to the package library
Divergent transitions > 10	Increase `adapt_delta` toward 0.99; inspect pairs plots with `bayesplot::mcmc_pairs()`
Low E-BFMI (< 0.2)	Check prior–likelihood conflict; consider supplying `fixed_a` for poorly identified lower asymptote
Max treedepth hits frequent	Increase `max_treedepth` in `fit_bayes_single()` or widen priors
Vignette fails during `R CMD check`	Add `eval = requireNamespace("curveRcore", quietly = TRUE)` to the setup chunk
`bead_assay_example` not found	Run `usethis::use_data(bead_assay_example)` and add a `R/data.R` roxygen block
Function missing from Reference	Check `@export` is present; re-run `devtools::document()`
LOO Pareto-k > 0.7	That calibrator is high-leverage; consider moment-matching via `loo::loo(moment_match = TRUE)`
`docs/` not served by GitHub	Settings → Pages: branch `main`, folder `/docs`

See vignette("bayesian-quickstart", package = "curveRbayes") for the full worked example.