Model Description Language (MDL) - a standard for model communication and interoperability
MK Smith (1), S Moodie (2), R Bizzotto (3), E Blaudez (4), E Borella (5), L Carrara (5), P Chan (1), M Chenel (6), E Comets (7), R Gieschke (8), K Harling (9), L Harnisch (1), N Hartung (10), A Hooker (9), MO Karlsson (9), R Kaye (11), C Kloft (10), N Kokash (12), M Lavielle (13), P Magni (5), A Mari (3), F Mentre (7), C Muselle (11), R Nordgren (9), H Nyberg (11), ZP Parra-Guillen (10), L Pasotti (5), N Rode-Kristensen (12), G Smith (13), MJ Swat (14), N Terranova (15), F Yvon (14), N Holford (9).
(1) Pfizer; (2) Eight Pillars Ltd; (3) Consiglio Nazionale delle Ricerche; (4) Lixoft; (5) Universita di Pavia; (6) Servier; (7) INSERM; (8) Roche; (9) Uppsala Universitet; (10) Freie Univeristät Berlin; (11) Mango Solutions; (12) Novo-Nordisk; (13) Cyprotex; (14) EMBL-EBI; (15) Merck.
Objectives: The DDMoRe Model Description Language (MDL)[1] and Pharmacometrics Markup Language (PharmML)[2] standards have been developed to convey information about models and tasks. MDL provides a means for modellers to describe and understand pharmacometric models irrespective of the tool used for analysis, while PharmML provides the basis for interoperability between software tools. The aim of this work is to show that MDL can be used to define models that are interoperable across modelling software tools and that are also easy to understand and share.
Methods: In developing MDL we have looked at features in established languages and adopted features that will facilitate interoperability, while retaining the flexibility to describe complex models. Having a well-defined software interchange standard (PharmML) and mapping MDL into PharmML allows us to focus on describing model features with one target in mind – PharmML. MDL conveys, in an accessible, user (analyst) friendly way, the models that can be encoded in PharmML. Converter tools then interpret the PharmML rather than the MDL for each software target. Testing this conversion and comparing output downstream allows us to check the translation.
Results: At the time of writing, MDL has been used to encode 10 Use Cases shared as part of the DDMoRe software installation describing common population pharmacokinetic model features and simple efficacy models with non-continuous outcomes. These models have been shown to be interoperable across commonly used NLME software: NONMEM, Monolix and simulx. The Design and Prior objects in MDL facilitate use with optimal design software such as PFIM and Bayesian software such as WinBUGS. Over 40 models have been encoded in MDL, published and shared via the DDMoRe Model Repository. These models cover disease areas such as oncology, CNS, infectious diseases. In excess of 60 delegates have been trained in MDL via the DDMoRe disease area training courses.
Conclusions: The interoperability of the Use Cases proves the desired outcome that a single model expressed in MDL can be used within a Pharmacometric workflow for a variety of tasks regardless of what target software is available to the user. Encoding models for the DDMoRe repository using MDL ensures that these models are readily understandable and shareable. Training pharmacometricians in MDL, a lingua franca, allows models and modelling concepts to be consistently defined, independent of software tools.
References:
[1] N Holford, MK Smith. MDL - The DDMoRe Modelling Description Language. PAGE 22 (2013) Abstr 2712 [www.page-meeting.org/?abstract=2712]
[2] MJ Swat et al. Pharmacometrics Markup Language (PharmML): Opening New Perspectives for Model Exchange in Drug Development. CPT: Pharmacometrics & Systems Pharmacology (2015) 4; 316-319. http://dx.doi.org/10.1002/psp4.57