Anne Kümmel (1), Benjamin Guiastrennec (1), Henning Schmidt (1)
(1) IntiQuan, GmbH
Background & Objectives
Pharmacometric analyses can influence decisions such as clinical trial design, drug approval, and labeling, as such, a high degree of traceability and reproducibility needs to be ensured. These steps are essential to enable collaborators, sponsors and regulatory agencies to properly evaluate the quality of and the validity of the results [1]. This can be challenging as these analyses, often complex, rely on multiple software for data processing (SAS, R), data analysis (R, NONMEM, Monolix, NLMIXR, PopED), and reporting which themselves have additional dependencies (compilers, libraries) [2,3].
IQRtools improved reproducibility and traceability with an R-based workflow from data preparation up to report [4]. Still, the problem of computational reproducibility persists, as differences in any third-party program can cause breaking changes and yield to a lack of confidence in the capacity to reproduce given results in the future. These issues are not specific to the pharmacometric field, and recently lead to the rise of containerized applications as supported by the docker project [5].
Herein we demonstrate the development and use of a docker image to create a reproducible pharmacometrics modeling environment that can be deployed on all standard operating systems.
Methods
The IQdesktop v0.2.0 docker image based on Ubuntu (v16.04.6 LTS) featured the standard pharmacometrician toolbox namely: R (v3.5.3 using a CRAN snapshot), Rstudio (v1.2), NONMEM (v7.4.3), PsN (4.9.0), Monolix (2019R1), NLMIXR (v1.1.1-2), popED (v0.4.0), IQRtools (v1.1.1), IQReport (v1.24) and git (v2.7.4) was developed. Each new version of IQdesktop is associated with a unique snapshot of specific tool versions, thus, to ensure computational reproducibility only the image version needs to be recorded. The IQdesktop image was validated upon its generation by running extensive script-based testing suites.
To demonstrate the benefits of computational reproducibility, a 1-compartment PK model with first order absorption was simulated up to 72h after a single dose and re-estimated in NONMEM 7.4.3 with FOCE and SAEM run on a single core under three 64-bit operating systems (Windows 10 Pro, macOS 10.14.6 and Ubuntu 16.04.6 LTS) using A) a standard installation and B) an IQdesktop environment.
Results
The IQdesktop image was successfully developed. From the validated image, “containers” (running instance of the image) were easily deployed on the different systems, the only requirement being a docker installation. The containers were accessed via a VNC or SSH client. The simulation-estimation exercise revealed that despite the same NONMEM version, data and models, with the local installation small differences (up to 3.2%) could be seen in the SAEM parameter estimates, whereas these differences were not present under the IQdesktop environment.
Conclusions
Setting up a validated system for pharmacometric analyses can be time consuming, requiring the installation and configuration of multiple programs which come multiple versions. With this work we have shown that despite using the same model, data and NONMEM version some differences in parameter estimates can occur across different computers system. With IQdesktop we established a validated and scalable pharmacometric modeling environment that was able to eliminate non-reproducibility due to differences in the computing environments.
References
[1] Ying C. Ou et al. Integration of Biostatistics and Pharmacometrics Computing Platforms for Efficient and Reproducible PK/PD Analysis: A Case Study. The Journal of Clinical Pharmacology. 53(11), 1112–1120, 2013.
[2] Justin J. Wilkins, Niclas Jonsson. Reproducible Pharmacometrics. Using Reproducible Research methodologies to improve pharmacometric analyses. PAGE meeting, 2013 [www.page-meeting.org/?abstract=2774].
[3] Reproducible Environments Manage environments for data science. environments.rstudio.com.
[4] IQRtools. Modeling & Simulation in R. iqrtools.intiquan.com.
[5] Carl Boettiger. An introduction to Docker for reproducible research, with examples from the R environment, ACM SIGOPS Operating Systems Review, Special Issue on Repeatability and Sharing of Experimental Artifacts. 49(1), 71-79, 2015.
Reference: PAGE () Abstr 9378 [www.page-meeting.org/?abstract=9378]
Poster: Methodology - Other topics