Patrick Nolain (1), Romain Combet (1), David Marchionni (1), Heiner Speth (2), Jean-Marie Martinez (1), David Fabre (1)
(1) Sanofi R&D, Montpellier, France (2) Sanofi R&D, Frankfurt, Germany
Introduction: Pharmacometrics workflow can be defined by a succession of well-defined tasks ranging from exploratory data analysis, modeling (execution, diagnostics, validation, simulation) and communication. Although each of these steps often require dedicated software and technologies, they share common needs (e.g. data visualization, reporting) which could take benefit from interactive analyses. The recent growth of data science related tools in R[1], in particular the web-application framework Shiny[2], brings new opportunities for the development of tools facilitating these activities.
Methods: A suite of interactive web-applications was developed to help the processing of several stages of a pharmacometric analysis using R and the shiny package.
Results: The PopkinR suite consists of four main components (already available):
- PMXplore: an exploratory data analysis application of NONMEM-like datasets
Provides interactive visualizations and summaries of various data types (dependent variables, dosing regimens, covariates) and dataset manipulation functionalities. - PMXrun: a NONMEM runs management application
Within the Sanofi computing cluster environment (Popkin[3]), pilots the execution of single (new or based on a prior run) or batch runs (e.g. initial values search, sensitivity analysis, covariates screening, bootstrap), monitors estimation convergence and supervises the cluster load in real time. - PMXploit: an R package for NONMEM post-processing analysis
The embedded shiny application allows pharmacometricians to interactively analyze NONMEM runs: obtain summaries of estimation results (e.g. convergence, parameter estimates, precision, shrinkage), visualize plots on the fly (e.g. observed data, goodness-of-fit, parameters and covariates distributions and correlations), compute numerical quality criteria, detect outliers, generate Visual Predictive Checks and compare multiple runs results. Generated plots and tables can easily be integrated into reports as well as splitted and filtered for subgroups analyses. - SimShiny: dynamic model-based simulations applications for communication of modeling results
Following the development of a model, a dedicated SimShiny application is designed to perform dynamic simulations and visualize model predictions. Model equations are implemented in R (using mrgsolve[4] or RxODE[5] packages) and user interfaces are adapted to each particular situation like comparison of dosing scenarios, computation of exposure parameters, comparison to models from the literature or exploration of complex dynamical systems. These tools facilitate communication of modeling results and foster collaboration with non-modelers, increasing the visibility of pharmacometry and its contribution to decision-making.
Conclusions: A suite of web-applications dedicated to several steps of pharmacometrics workflow, from data exploration to simulation-based decision making, was developed. PopkinR shows the ability of interactive applications to improve the efficiency of pharmacometricians’ work and the communication of modeling and simulation contribution to a wider audience to support drug development. Two additional components (a non-compartmental analysis and a therapeutic drug monitoring applications) are in development to complete the PopKinR suite.
References:
[1] R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/
[2] Winston Chang, Joe Cheng, JJ Allaire, Yihui Xie and Jonathan McPherson (2017). shiny: Web Application Framework for R. R package version 1.0.5. https://CRAN.R-project.org/package=shiny
[3] Speth H (2004). A Linux cluster for population pharmacokinetic analyses. International journal of clinical pharmacology and therapeutics 2004 42:3 (189-190)
[4] Kyle T Baron (2017). mrgsolve: Simulate from ODE-Based Population PK/PD and Systems Pharmacology Models. R package version 0.8.10. https://CRAN.R-project.org/package=mrgsolve
[5] Matthew L. Fidler, Melissa Hallow and Wenping Wang (2017). RxODE: Facilities for Simulating from ODE-Based Models. R package version 0.6-1. https://CRAN.R-project.org/package=RxODE
Reference: PAGE 27 (2018) Abstr 8684 [www.page-meeting.org/?abstract=8684]
Poster: Methodology - Other topics