predmicror provides predictive microbiology model functions and convenience wrappers for fitting primary growth, microbial inactivation, and cardinal parameter models to experimental data. It also supports omnibus (nonlinear mixed-effects) and dynamic (ODE-based time-varying) modelling.

The package can be used in two complementary ways:

call the exported model equations directly inside gslnls::gsl_nls();
use the higher-level fit_*() wrappers for routine fitting, diagnostics, and model comparison.

Installation

install.packages("devtools")
devtools::install_github("fsqanalytics/predmicror")

Model catalogue

List all supported models with:

library(predmicror)
predmicror_models()
predmicror_models("growth")
predmicror_models("inactivation")
predmicror_models("cardinal")

Workflow	Wrapper	Response scale	Examples
Primary growth	`fit_growth()`	natural log (`lnN`)	`HuangFM`, `BaranyiFM`, `RossoFM`
Microbial inactivation	`fit_inactivation()`	base-10 log (`logN`)	`WeibullM`, `GeeraerdST`, `WeibullPH`
Cardinal parameters	`fit_cardinal()`	sqrt growth rate (`sqrtGR`)	`CMTI`, `CMPH`, `CMAW`, `CMInh`
Dynamic (time-varying)	`fit_dynamic_growth/inactivation()`	natural log (`lnN`)	Huang ODE, Weibull-Peleg ODE
Omnibus (mixed-effects)	`fit_omnibus_growth/inactivation()`	per primary model	any primary + secondary covariates

Quick start: growth

library(predmicror)
data(growthfull)
fit <- fit_growth(
  data = growthfull,
  model = "HuangFM",
  time = "Time",
  response = "lnN",
  start = list(Y0 = 0, Ymax = 22, MUmax = 1.7, lag = 5)
)
summary(fit)
coef(fit)
plot(fit)

Inactivation

data(bixina)
fit <- fit_inactivation(
  data = bixina,
  model = "WeibullM",
  time = "Time",
  response = "lnN",
  start = list(Y0 = 5.6, sigma = 12, alpha = 1)
)
fit_metrics(fit)
plot(fit)

Cardinal parameter model

data(salmonella)
fit <- fit_cardinal(
  data = salmonella,
  model = "CMTI",
  x = Temp,
  response = sqrtGR,
  start = list(Tmax = 42, Tmin = 1, MUopt = 1, Topt = 37)
)
coef(fit)

Model comparison

huang <- fit_growth(growthfull, model = "HuangFM", time = "Time",
  response = "lnN", start = list(Y0 = 0, Ymax = 22, MUmax = 1.7, lag = 5))
baranyi <- fit_growth(growthfull, model = "BaranyiFM", time = "Time",
  response = "lnN", start = list(Y0 = 0, Ymax = 22, MUmax = 1.7, lag = 5))
compare_models(HuangFM = huang, BaranyiFM = baranyi, sort_by = "AIC")

Dynamic (time-varying environment)

profile <- dynamic_profile(
  time = c(0, 5, 10),
  temperature = c(12, 20, 25)
)
pred <- predict_dynamic_growth(
  profile = profile,
  start = list(logN0 = 2, logNmax = 8, a = 0.08, Tmin = 5, lag = 1),
  times = seq(0, 10, length.out = 50)
)
plot(pred$time, pred$logN, type = "l")

Omnibus (mixed-effects)

# Each group (Condition) has its own curve
dat <- data.frame(
  Condition = rep(1:3, each = 7),
  Time = rep(seq(0, 6), 3),
  lnN = HuangNLM(rep(seq(0, 6), 3), Y0 = 2, Ymax = 8, MUmax = 0.6)
)
fit <- fit_omnibus_growth(
  data = dat,
  primary = "HuangNLM",
  time = "Time",
  response = "lnN",
  group = "Condition",
  random = Y0 ~ 1,
  start = c(Y0 = 2, Ymax = 8, MUmax = 0.6)
)

Response scale conventions

The fitting wrappers do not transform the response automatically:

Growth models: natural logarithm of concentration (lnN)
Inactivation models: base-10 logarithm (logN)
Cardinal models: square root of growth rate (sqrtGR)
Dynamic models: natural logarithm (lnN)

Documentation

Full reference and vignettes:

https://fsqanalytics.github.io/predmicror/

Citation

citation("predmicror")

Bugs

https://github.com/fsqanalytics/predmicror/issues