PAGE. Abstracts of the Annual Meeting of the Population Approach Group in Europe.
PAGE 25 (2016) Abstr 5792 [www.page-meeting.org/?abstract=5792]
Poster: Methodology - Other topics
Jonathan Chard (1), Justin Wilkins (2), Amy Cheung (3), Evan Wang (4), Mike K Smith (5), Phylinda Chan (5), Gareth Smith (6), Richard Kaye (1), Maria Luisa Sardu (7), Stuart Moodie (8)
(1) Mango Solutions, (2) Occams, (3) Astra Zeneca, (4) Eli Lilly, (5) Pfizer, (6) Cyprotex, (7) Merck Serono, (8) Eight Pillars Ltd
Objectives: To develop a standard, implemented with a workflow software tool, for capturing the full range of activities and entities that are performed during a pharmacometric analysis, based on existing standards. Capturing the provenance of task outputs (how was this created) as well as providing knowledge management for the pharmacometrics workflow (how did we get to this model) facilitates reproducibility, sharing, and communication of results with others. Using this standard, we can visualise the steps taken during the analysis, reproduce analysis steps, and capture decisions, assumptions, key steps, and support the process of quality control, as suggested in the definition of Model-Informed Drug Discovery and Development (MID3)
Methods: Several existing workflow tools and provenance capture standards were evaluated [3,4,5], but the PROV-O ontology was selected due to its wide adoption, extensibility and suitability for capturing the provenance and relationships between activities and entities within and across projects. Analysis artefacts, actions, information and relationships were mapped onto concepts defined within PROV-O. Tools were developed to support the pharmacometric workflow; storing files in Git , generating the provenance information representing the steps taken by the pharmacometrician, and to query the captured information to visualise, report, and regenerate the artefacts within an analysis.
Results: The standard allows tracking of users, software tools, and files in an analysis, while capturing assumptions, decisions and relationships extending beyond input to output. Information can be captured at multiple levels of detail, allowing a reviewer to understand key decisions taken during an analysis, or to trace through the software used to generate results. It is possible to identify project artefacts that are out of date (e.g. a diagnostic plot that should be recreated due to dataset change), and re-run activities. Analysts can apply this information to generate documentation, from run records to complete reports. Knowledge shared between team members is enhanced, avoiding duplication of work, increasing quality and reproducibility. Traceability assists reviewers and regulators to evaluate assumptions, results and conclusions.
Conclusions: Capturing structured information with software tools helps to ensure data integrity, facilitating QC and adoption of MID3 concepts.
Acknowledgements: This work is presented on behalf of the DDMoRe project (www.ddmore.eu).