Programming languages and software are critical tools in the field of pharmacometrics, an area that has grown significantly in recent years. However, the rapid development of new methods and modelling software has presented challenges for researchers. Because so many of these new tools are built without a standardised language or formulation method, variation can make it difficult to incorporate them into existing workflows readily. Manual translation of the underlying pharmacometric model can be one solution, but this is often time consuming and may introduce glitches, either through translational failure or human error.
The need for a unified exchange model was recognized several years ago by the NonLinearMixed Effects (NLME) Consortium, who undertook drafting an initial format specification. Development stalled after this initial interest, but in 2011 these early-stage plans served as a starting point when the European Innovative Medicines Initiative (IMI) incepted the Drug Disease Model Resources (DDMoRe) Consortium. Specifically, this project has set out to “build and maintain a universally applicable, open source, model-based framework, intended as the gold standard for future collaborative drug and disease Modeling & Simulation.” To this end, DDMoRe have developed the Pharmacometrics Markup Language (PharmML) to integrate existing tools such as MatLab, SIMCYP Simulator, R, NONMEM, and others. Fields such as neuroscience and systems biology, already have standard model exchange protocols in place that have significantly changed the course of biological modelling in these areas. However, pharmacometrics presents the unique challenge of developing a standard within an existing network comprised of various approaches for model encoding.
Two popular tools (NONMEM and Monolix) are built with distinct languages and different functional styles. NONMEM is assignment-based, which is very powerful and offers a high degree of flexibility at the cost of standardisation capabilities. Monolix on the other hand is declarative and rigidly structured with explicit syntactical protocol that makes standardisation an easier process, but it lacks flexibility. Both approaches are valid and important to pharmacometric modelling, but differences in formatting and layout make conversion and unification a complex challenge.
Another obstacle is that the field of pharmacometrics has a very broad scope. Whereas many other areas of drug development specialise in a single organ or system, pharmacometrics does not and can range from models of intracellular pathways to a whole organisms.
If these challenges are to be overcome and the goal of creating a unified exchange that functions across workflows is to be met, development of a single format for unambiguous model formulation is the critical first step.
PharmML is based on XML, a popular markup language, which offers several advantages over other encoding protocols. First, by definition, XML is extensible (XML = Extensible Markup Language), meaning new syntax and vocabulary can be added over time, allowing it to grow to meet new demands. It is also a broadly used format, so it can interact with other schemas based on the XML format. Finally, because it is so broadly used, there are many tools the support XML-based processing and development.
With XML as a framework, PharmML can encode Model Definition, Trial Design, and Modelling Steps as described below:
At PharmML’s core is the Model Definition, which was created on the explicit mathematical structure of NLME’s first drafts. The Model Definition itself houses five submodels, each with a distinct structure and hierarchy to accommodate complex mathematical techniques such as matrices, linear and nonlinear covariates, algebraic and differential equations, and allometric scaling within a broad spectrum of parameters and possible formulations.
The Trial Design section of PharmML improves upon the traditional approach where the trial design is executed within a dataset. In PharmML, this section encodes both simulation and optimal design tasks in addition to functionality for estimation tasks, formulating a study design without dataset dependency.
The third section, Modelling Steps, houses the definitions of the basic tasks a model is designed to perform. Estimation and simulation are the two types of basic tasks currently supported by PharmML, but because of its extensible nature from the XML format, more may be added at a later date.
PharmML v. 0.6, the most recent iteration, was released in January 2015 and new elements are still in development. The Standardized Output (SO) element will be a storage format for model results. Optimal Experimental Design (OED) enhances the Trial Design section by allowing for additional design space for domain-specific optimisation settings. Support for SBML-coded structural models is also expected to be forthcoming.
Hopefully, with each subsequent release, PharmML will see wider adoption, streamlining varied and complex workflows, standardising definitions and therefore facilitating reporting and reproducibility of research and encouraging the development of new tools and software within a unified format.
An overview of PharmML is included in the following recent publication:
Swat MJ et al., (2015) Pharmacometrics Markup Language (PharmML): Opening New Perspectives for Model Exchange in Drug Development. CPT Pharmacometrics Syst Pharmacol 4; 316-319.