(Note on 2023-11-03: If you would like to fit thermal performance curves using an R package, I recommend that you use the rTPC package by Daniel Padfield and Hannah O’Sullivan (available on CRAN) rather than thermPerf. rTPC is much more advanced than thermPerf and it has very nice documentation in the form of detailed vignettes.)
Disclaimer: This is a collection of functions that we use in our research group to fit curves to bacterial growth and yield data. The defaults in these functions are tuned for curve fitting using this data, and might be inappropriate for other temperature ranges/traits values/curve shapes!
Disclaimer #2: Note that the AIC calculation might not be perfectly accurate; see the warning displayed when the package is loaded:
*** WARNING: there is some concern about the way logLik is calculated by R for a nls fit (see http://r.789695.n4.nabble.com/LogLik-of-nls-td4657207.html). Maybe it would be better to calculate it "by hand" from the nlsLM fit? ***
library(devtools)
install_github("mdjbru-R-packages/thermPerf")library(thermPerf)
# Some thermal performance data
temp = c(17, 21, 24, 28, 31, 33,
17, 21, 24, 28, 31, 33)
growth = c(0.3, 0.4, 0.68, 0.82, 0.78, 0.3,
0.4, 0.45, 0.58, 0.75, 0.83, 0.6)
# Plot the raw data
plot(temp, growth, pch = 21, bg = "grey", las = 1,
xlim = c(15, 35), ylim = c(0, 1), bty = "n")
# Fit the default models and plot the results
fits = fitModels(getModelLibrary(), temp, growth)
plot(fits, xlim = c(15, 35), ylim = c(0, 1), las = 1)
# Get the AIC weights and visualize them
weights = calculateAIC(fits)
plot(weights)
# Interpolate trait value for some temperature
interpolateAtTemp(fits, 26)Note that the AIC calculation might not be perfectly accurate; see the warning displayed when the package is loaded:
*** WARNING: there is some concern about the way logLik is calculated by R for a nls fit (see http://r.789695.n4.nabble.com/LogLik-of-nls-td4657207.html). Maybe it would be better to calculate it "by hand" from the nlsLM fit? ***
Models to be fitted are stored in objects of class customModel. One can
define his own models to be fitted to the data, or use the (small) library of
models already available in the library. Note: the built-in library is very
small and contains only models of interest for our own research project. You
should probably check which of those models are pertinent for your project, and
add more models as suits your needs.
# Available, pre-built models are available from getModelLibrary()
library(thermPerf)
models = getModelLibrary() # This returns a list of customModel objects
str(models, 1) # There are 10 models in the pre-built library
## List of 10
## $ linearFit :List of 5
## ..- attr(*, "class")= chr "customModel"
## $ briere1 :List of 5
## ..- attr(*, "class")= chr "customModel"
## $ briere2 :List of 5
## ..- attr(*, "class")= chr "customModel"
## $ lactin1 :List of 5
## ..- attr(*, "class")= chr "customModel"
## $ candidate01:List of 5
## ..- attr(*, "class")= chr "customModel"
## $ candidate02:List of 5
## ..- attr(*, "class")= chr "customModel"
## $ candidate03:List of 5
## ..- attr(*, "class")= chr "customModel"
## $ candidate04:List of 5
## ..- attr(*, "class")= chr "customModel"
## $ candidate05:List of 5
## ..- attr(*, "class")= chr "customModel"
## $ candidate06:List of 5
## ..- attr(*, "class")= chr "customModel"library(thermPerf)
models = getModelLibrary() # This returns a list of customModel objects
myModel = models[["briere1"]] # We take the list element called "briere1"
myModel # This will output a summary of the customModel object
## Custom model: Briere 1
##
## model formula: y ~ a * x * (x - t0) * (tmax - x)^(1/2)
## parameters (n=3): a, t0, tmaxEach model is actually a list with all the information needed to perform a fit:
str(myModel)
## List of 5
## $ function :function (x, params)
## $ name : chr "Briere 1"
## $ formula :Class 'formula' length 3 y ~ a * x * (x - t0) * (tmax - x)^(1/2)
## .. ..- attr(*, ".Environment")=<environment: 0x3011878>
## $ parameters: chr [1:3] "a" "t0" "tmax"
## $ starting :List of 3
## ..$ a : num 1
## ..$ t0 : num 15
## ..$ tmax: num 35
## - attr(*, "class")= chr "customModel"The model elements are:
- a function that can evaluate the model at an x value (i.e. a temperature value) given a set of parameter values for the model
- a name for the model
- a formula describing the model
- the model parameter names
- a list of starting values for the model parameters, to be used when the fit
is performed with the
minpack.lm::nlsLMfunction.
Custom models can easily be defined using the buildModel function:
library(thermPerf)
# Create a model for third-order polynomial
# 1) Function
mFunction = function(x, params) {
# params model parameters, a0, a1, a2, a3
a0 = params[["a0"]]
a1 = params[["a1"]]
a2 = params[["a2"]]
a3 = params[["a3"]]
return(a0 + a1 * x + a2 * x^2 + a3 * x^3)
}
# 2) Name
mName = "3rd-order polynomial"
# 3) Formula
mFormula = y ~ a0 + a1 * x + a2 * x^2 + a3 * x^3
# 4) Model parameters
mParams = c("a0", "a1", "a2", "a3")
# 5) List of starting values for the parameters
mStarting = list(a0 = 0, a1 = 1, a2 = 0.5, a3 = 0.1)
# Create the customModel object
myModel = buildModel(mFunction, mName, mFormula, mParams, mStarting)
# Summary
myModel
## Custom model: 3rd-order polynomial
##
## model formula: y ~ a0 + a1 * x + a2 * x^2 + a3 * x^3
## parameters (n=4): a0, a1, a2, a3The model can be used for fit now:
# Some thermal performance data
temp = c(17, 21, 24, 28, 31, 33,
17, 21, 24, 28, 31, 33)
growth = c(0.3, 0.4, 0.68, 0.82, 0.78, 0.3,
0.4, 0.45, 0.58, 0.75, 0.83, 0.6)
# Plot the raw data
plot(temp, growth, pch = 21, bg = "grey", las = 1,
xlim = c(15, 35), ylim = c(0, 1), bty = "n")
# Fit the model, along with some models from the model library
models = getModelLibrary()[c("linearFit", "lactin1")]
models[["myModel"]] = myModel
fits = fitModels(models, temp, growth)
plot(fits)