NLME Modeling in Shiny
Non-linear Mixed Effect (NLME) Modeling is a powerful tools for the analysis of population data. The use of typical NLME modeling tools, such as NONMEM and MONOLIX, requires specialized knowledge that, especially in the case of NONMEM, puts more focus on tools specifics than on the task a modeler actually would like to perform. In every day modeling & simulation life standard analyses are recurring that would benefit from more easy access to NLME modeling than having to wait for a modeler to be available, a custom dataset programmed, and a report written.
- A clinical pharmacologist who receives first batches of single or multiple dose First-in-Human PK data and would like to conduct a quick PK modeling exercise to assess the PK properties of the compound in humans that are more reliable than an NCA
- A DMPK scientist wanting to extrapolate preclinical PK information to human
What was missing was a graphical user interface, allowing to efficiently define models, data, and perform NLME modeling analyses. More importantly, such an interface should be able to handle a number of different NLME parameter estimation tools, such that the user can select the tool that is most appropriate for the available data. In addition to the standard tools NONMEM and MONOLIX, also open-source tools should be allowed to use, limiting the licensing issues and costs.
Our approach was to use R Shiny to encapsulate the underlying complexity (data/model handling, simulation, conversion of model to the syntax required by the selected NLME modeling tool, parameter estimation, diagnostic plot generation) in a user-friendly graphical user interface. The underlying functionality has been developed in R and is made available as the IntiQuan R Tools.
The required data format has been simplified to allow anyone with basic modeling insights to create the dataset for their data. As NLME parameter estimation tools NONMEM, MONOLIX, and NLMIXR have been interfaced and the user can select either of them without requiring training in the different NLME tools.
Models are run in the background and results are displayed in the Shiny App, comprising a parameter results table with true parameter names, standard errors, shrinkage information, diagnostic plots, and the executed model control stream.
The Shiny App has been developed to allow multiple NLME modeling projects being handled at the same time. Each modeling project in this context is associated with a specific dataset.
When run locally on a computer, models and results can be accessed from a nicely organized folder based structure.
The work resulted in a user-friendly Shiny Interface for NLME modeling. The main user group for this Interface are NLME modeling knowledgeable DMPK or ClinPharm scientist who do not want to write lengthy and potentially cryptic NLME modeling tool code and do not want to wait for a modeler to have time to analyze their data. The model library, designed into the Shiny App, can hold all typical PK models of typical interest, requiring the user to only upload the data and get a first model based analysis.
The NLME Shiny Interface is available online and for download
- See it online: >>>here<<<
Note that estimation is disabled for NONMEM and MONOLIX due to licensing reasons. NLMIXR is not compatible with running over a Shiny Server. To perform parameter estimation with this Shiny App, you will need to get the offline version (see below).
- How to get the offline version: