Next-Generation Modeling Language
Mike Dunlavey
Pharsight Corp.
Objectives: Develop new language-based tools for modeling and simulation.
Methods: A new modeling language has been developed by Pharsight and is currently being tested together with model function evaluation, NLME estimation, and trial simulation algorithms. It is designed to be a fairly easy transition from NONMEM®, but have a more modern structure and provide direct language support for expressing models with multiple responses, multiple administration routes, non-Gaussian responses (categorical, time-to-event, and counting-process), gaussian responses with quantification limits, and differential equations. Arbitrary variable names for fixed and random effects and free-form syntax are used. A full set of simulation and table-generation capabilities is also supported. The language is designed to make it easy to incorporate fitted models into subsequent trial design simulations, as well as to facilitate covariate selection and optimization procedures.
Supported NLME algorithms include FO, FOCE, Laplacian, and a sophisticated nonparametric method based on modern primal-dual optimization methods. The model function evaluation algorithm for models expressed with differential equations can automatically recognize systems for which matrix exponentiation is appropriate. All NLME algorithms have been parallelized using MPI for multiprocessor execution of single jobs.
The syntax is designed so that it can be used stand-alone or in concert with scripts in the S-PLUS® language.
Results: Example code will be displayed.
