Sahota, Tarjinder (1), Ruau, David (2), Fox, Robert (3), Al-Huniti, Nidal (3)
(1) AstraZeneca, Quantitative Clinical Pharmacology, IMED, Cambridge, UK (2) AstraZeneca, Scientific Computing Solutions, GMD, Cambridge, UK (3) AstraZeneca, Quantitative Clinical Pharmacology, IMED, Gaithersburg, US.
Objectives:
Knowledge management is the ability of an organisation to retain, reuse and share knowledge when/where it is needed. Have you ever spent time implementing model code that you later found out had already been implemented by one of your colleagues? Perhaps you missed that colleague’s email or presentation. Have you ever had to continue a colleague’s work but struggled to figure out what he/she did and in what order? Have you ever had trouble reproducing results you produced mere months ago because now there’s a package/R version incompatibility? Have you ever had to repeat entire model development steps because of a small change to the initial dataset? We aim to develop a suite of R packages to overcome these challenges.
Methods:
Two R packages have been created:
TidyProject: Directory and code management.
NMproject: A light interface to NONMEM from R for script based model development (requires installation of NONMEM and Perl speaks NONMEM (PsN)).
Installing NMproject will install both packages. They will soon be released on the Comprehensive R Archive Network (CRAN), but are currently available on GitHub via the following R command: devtools::install_github(“tsahota/NMproject”). NMproject can be configure for a wide variety of infrastructures.
Results:
AstraZeneca employees have successfully used NMproject to conduct NONMEM model development for over two years. A short summary of features of features is available on Youtube: https://www.youtube.com/watch?v=b7oBb6QZub8. Features include:
Tidy directory structure: Standardised, version controlled, directory structure for all NONMEM/R users. Functions to check your own adherence to best practices.
Run database: A hidden SQLite database monitors all runs, ensures new outputs don’t overwrite existing outputs and enables easy creation of run histories.
Code library: An extensive library of templated control streams utilising best coding practices to reduce time spent coding NONMEM control streams, maximise stability and reduce errors. Create new templates for private use or for your organisation. Interface to search using keywords, tags, or raw text.
Private project library: Long term reproducibility and consistent running of R scripts between users. Store session information in case of catastrophic failure.
Script based model development: Code your model development process using end-to-end R scripts. Switch between interactive (for normal model development) and non-interactive mode (for sequencing multiple runs and post processing steps).
Shiny run monitor: Shiny GUI interface for run monitoring and results comparison.
Conclusions:
NMproject is a prototype of industrialised knowledge management for NONMEM and R users.
Reference: PAGE 27 (2018) Abstr 8650 [www.page-meeting.org/?abstract=8650]
Poster: Software Demonstration