# residual error and bio-availability can be added in the des_func e.g.
CP <- (A2/V) * F1 * (1 + rnorm(length(A2),0,sqrt(0.01)))
# lag time should be adapted in the events data frame e.g.
events <- dose_func(cmt=1,value=300,ndose=3,tau=24)
events$time <- events$time + pars['TLAG']
# infusions should be handled in the event and are supported in dose_func
# see template models how this can be adapted in the model part (?tmpl_model)
dose_func(cmt=4,value=300,ndose=3,tau=24, tinf=2)Translating model syntax from one language to another can become quite complex. This package will try to do this as good as possible, however there might be times that things just do not work out of the box. The most important things to take into account are listed here. If you find some situations that are not handled correctly or have some ideas how it could work please file an issue on github.
Model types
There are different types of models that can be used in NONMEM. Currently the package mainly supports models that are written in differential equations (e.g $DES block). Models written without differential equations but still user defined (e.g. $PRED block) can be translated but likely need further inspection. For these types of models, the deSolve package is not really suited. The package will translate it to an R function but it is advised to simplify the result and use base R for simulations. For rxode2 and mrgsolve there are more options for simulating these types of models. Likely less adaptations are necessary, but it is advised to read the additional documentation regarding these models once translated for rxode2 and/or mrgsolve.
Furthermore within NONMEM there are multiple ADVAN routines for common PK models. The package will not translate models defined using the general ADVANS (e.g. models without $DES block). The main reason for this is that translating these types of models is generally fast and easy. Also the rxode2 and mrgsolve packages both have options to handle these types of models. To simulate these types of models in deSolve or directly in base R, the package provide template models (see ?tmpl_model). When such a model is used in the convert_nonmem function, you would typically get the results for the blocks that can be translated, but the model will likely not compile. IT can be helpful though to translate parts that can easily be added to the the default models.
Lag time, bio-availability, residual error and infusions
Because both rxode2 and mrgsolve are focused towards pharmacometric models, all specifics regarding lag time, bio-availability, etc. should be handled correctly. However when working with deSolve there might be some considerations that should be take into account. When a bio-availability parameter is available, you should check if this parameter should be adjust in the model or should be added to the dosing part. Similar for lag time, this can be added in the dosing part by delaying all given doses by the lag time. Residual error can be added manually quite easily within the model or afterwards. With respect to IV infusions, this is something that should be taken into account in both the event dataset and within the model. The event dataset should define a rate of the infusion while in the model an additional compartment should be included. Some examples on how to handle these specifics when working with deSolve are given below:
Specific NONMEM coding
Within NONMEM there are different ways of coding certain blocks. For instance there are many ways of defining initial estimates for and values. The package supports most common coding trough the wonderful NMdata package but maybe not all. It is therefore advised not to use very specific coding here (e.g. defining as (value)xn to repeat estimates). For values, blocks are supported but using the ‘block same’ structure to code inter-occasion variability might need some double checking within the simulations. Also structures defining correlations instead of covariance will likely need adaptations.
Within NONMEM it is also possible to write abbreviated FORTRAN code (e.g. including things like COM(n) and/or $ABBREVIATED). Additionally, the usage of protected functions in R could lead to possible issues in the translated model (e.g. PROTECT option in $ABBREVIATED). This part is currently not supported by the package. In some cases the code is commented out in the translated models, however in certain cases it might be necessary to deleted/rewrite before translating the model.
The most important reserved keywords are taken into account. However given the amount of reserved keywords (not only in NONMEM but in the simulation packages as well), it could be that certain keywords are not supported. Please file an issue if you stumble upon an example that does not work.
General
Please be aware that the package reduce the need of manual translating. It is however advised to always carefully go through your simulation model to see if everything works as expected. Some parts are difficult or nearly impossible to translate. However if you have the idea that things can be implemented or do not work as expected you can always file an issue on github and I will try to handle it as good as possible.