Awesome

• petab-handling itself, fitting and simulation tool agnostic
  • https://github.com/PEtab-dev/PEtab
  • https://petab.readthedocs.io/
• dMod interface
  • https://github.com/dkaschek/dMod

Installation

# In ubuntu, it might be necessary to run this in bash
# sudo apt-get install libxml2-dev

if (Sys.info()["sysname"] == "Windows") {
  options(download.file.method = "wininet")
  devtools::install_url("https://sourceforge.net/projects/sbml/files/libsbml/5.18.0/stable/Windows/R%20interface/libSBML_5.18.0.zip")
} else {
  devtools::install_url("https://sourceforge.net/projects/sbml/files/libsbml/5.18.0/stable/R%20interface/libSBML_5.18.0.tar.gz")
}
devtools::install_github("dkaschek/dMod"   )
devtools::install_github("dlill/petab")

Notable functions (not exhaustive)

Learning

• petab_exampleList - Show all available examples
• petab_exampleRead - Load an example to play around with it or navigate to the folder including some R-scripts e.g. on Ubuntu+Gnome with system(paste0("nautilus ", petab_examplePath(exampleName = "01", "dir")), wait = FALSE)

Basic functionality for petab handling

• petab_mutateDCO - consistently manipulate experimentalCondition, observables and measurementData at once in very expressive data.table language. Everything, i and j can do in a data.table, petab_mutateDCO can to it as well!
• petab_plotData - Plot your data with great liberty in customizing the plot
  • ggCallbacks, aeslist
    • Easily customize plot in terms of layers, aesthetics, layout and labels, scales
  • Multipage output via ggforce::facet_grid/wrap_paginate
  • Asynchronous output via the future-package, so you don't wait forever until you can use your R-session again
  • Observable transformations on/off
• petab_python_setup Access the original petab development library. Sets up a virtualenv via the reticulate-package, to have a clean installation.

dMod functionality

• importPEtabSBML_indiv - Import PEtab to dMod. Can check if the input PEtab is changed to only recompile the model when necessary
• writePetab - If the model part is specified in dMod language, the SBML is exported
• pd_predictAndPlot2 - Simulate and plot your model
  • ggCallbacks, aeslist
    • Easily customize the ggplot in terms of layers, aesthetics, layout, labels, scales and theme
  • Multipage output via ggforce::facet_grid/wrap_paginate
  • Asynchronous output
  • Quick subsetting
    • Plot all internal states yes/no
    • Plot all conditions yes/no
    • Plot up to maximum time or match to data
  • Detailed subsetting
    • i-argument (in data.table parlance) to perform any kind of subset: e.g. subset on time, observableId, conditionId or a combination of them
  • Observable transformations on/off
  • Parameter options
    • If you have multistart fits, you can directly compare different model trajectories at different optima
    • If you have profiles or prediction profiles, you can use the left and right endpoints of all profiles for predictions which are then summarized to produce model uncertainty bands surrounding the prediction at the best fit

Contributing

Some thoughts on coding-style

• Code is debugged more often than it is written (similar to Python's philosophy that code is read more often than it is written). Therefore, code should be debuggable
  • Only few calls should span multiple lines (e.g. lapplys), and don't use pipes. Make every step explicit, so people can easily see intermediate results
  • No large functions of more than 100 lines, no if statements where you have to scroll to see the closing bracket. All of these can be refactored to use smaller helper functions which improve clarity of the progam's logic and help to debug by being more informative in the traceback.
  • Informative error messages, telling what went wrong, instead of telling that something went wrong
• Documentation WITH examples is a must
• Declare function imports properly, so functions and their roxygen skeletons can be moved to a different package easily
• Try to use git commit --amend and devtools::document until the code definitely works before pushing to keep the git log clean
• Reuse the semantics of some flags, e.g. NFLAGsave = 3 should always mean "check if input files changed, and if not, reload the previously saved version, if it exists"
• Try to program with good taste: https://www.youtube.com/watch?v=o8NPllzkFhE
• Optimally, push to a branch and create PR

Version history

• New in 0.3.1
  • Started resturucturing the reading of fit results
  • objective functions are moved to pd$objfns
  • be more strict when it comes to interpreting the petab: Use the objective prior as specified in fitting etc...
  • pd_cluster_* functions to run stuff on cluster
• New in 0.1.1
  • parameterFormulaInjection in petab$meta
    • Allows to specify arbitrary parameter transformations after the estimation parameter trafo and before the dynamic model
    • The new prd0 looks like gxp1*p0 where p0 is the trafo for est-scales and p1 is the "injected trafo", e.g. a steady state trafo
    • This trafo is the same for all conditions to stay consistent in the indiv-framework

References

R logo https://www.r-project.org/logo/ License https://creativecommons.org/licenses/by-sa/4.0/