PharmML – An Exchange Standard for Models in Pharmacometrics
Maciej Swat (1), Stuart Moodie (1), Niels Rode Kristensen (2), Nicolas Le Novère (3) on behalf of DDMoRe WP4 contributors.
(1) EMBL-EBI, Hinxton UK, (2) Novo Nordisk, Denmark, (3) Babraham Institute, UK
Objectives: A long-standing problem in Pharmacometrics is the lack of a common standard allowing for exchangeability of models between existing software tools, such as Bugs, Monolix, NONMEM and others. PharmML, as part of the DDMoRe interoperability platform [1], presented here in its 1st specification tries to fill this gap. The modelling framework is that of Nonlinear Mixed Effects Models, NLME, which allows for nonlinear models with random and fixed effects. This new standard provides encoding platform for approaches currently in use but also attempts to create support for novel elements.
Methods: The development of PharmML is based on requirements provided by the DDMoRe community, including numerous academic and EFPIA partners, use cases for various estimation and simulations tasks encoded in languages such as NMTRAN and MLXTRAN and mathematical documents outlining the statistical background prepared within DDMoRe ([2],[3]). The methodology included analysis of use cases and their implementation in Matlab, abstraction of the information necessary to encode a particular type of model, creating an XML schema and testing its performance and functionality. We reuse where possible existing standards, such as SBML to encode the structural model.
Results: The current specification supports the exchange of continuous models, in the form of algebraic equations or systems of ODEs. The parameter model offers a very flexible structure allowing for the use of most common parameters. It is linear in the transformed parameter, and the resulting additive structure allows for easy interpretation and implementation of its components, such as continuous and discrete covariates, correlation structure of the random effects and virtually any level of variability as nested hierarchy. The clinical trial model provides the modeller with almost unlimited possibilities to construct arbitrary study designs using only few basic building blocks, such as Treatment, Treatment Epoch and Group. PharmML is providing a means to annotate an arbitrary element of the model, making effective searching and reasoning on models in a repository possible. Models encoded in this way can be used not only for the standard tasks, such as simulation and estimation but also modelling and exploration.
Conclusions: This specification provides a solid basis for further development of PharmML in future. Subsequent releases will support discrete models, Bayesian inference framework, stochastic differential equations and Hidden Markov Models etc.
This work is presented on behalf of the DDMoRe project.
References:
[1] The DDMoRe project, www.ddmore.eu
[2] Lavielle, M. (June 25, 2012). Mathematical description of mixed effects models. Technical report, INRIA Saclay.
[3] Keizer, R. and Karlsson, M. (2011). Stochastic models. Technical report, Uppsala Pharmacometrics Research Group, Uppsala.
