Lars Lindbom, Pontus Pihlgren and E. Niclas Jonsson
Div of Pharmacokinetics and Drug therapy, Dept Pharmaceutical Biosciences,
Objectives: Many computer intensive methods useful in model building and model validation have been described in the literature [1, 2]. Their uses in population pharmacokinetics and pharmacodynamics have however been limited due to the lack of computing power and software. One way of addressing the limited computer power problem is to use distributed computing in the form of clusters or grids. This in turn requires software that supports parallel execution within a distributed computing environment. Perl-speaks-NONMEM (PsN)[3] is a programming library previously written by the authors to facilitate the development of software around the mixed effect modelling program NONMEM. Here we describe PsN-toolkit, a collection of computer intensive methods developed using PsN, intended for use in population pharmacokinetic and pharmacodynamic modelling.
Methods: The programming language Perl and the programming library PsN were used to implement the computer intensive methods. Support for parallel execution of the methods was constructed using the Perl module Parallel::ForkManager[4]. The object oriented structure of PsN was retained in the PsN toolkit methods. Nested dependencies between method classes were permitted, allowing for general inter-method communication.
Results: A collection of computer intensive methods intended for use in population pharmacokinetic and pharmacodynamic modelling using NONMEM has been constructed. Included methods are for example the Bootstrap, Cross Validation and a Stepwise Covariate Model Building tool. The methods share a common structure for preparation of data and model files, execution of NONMEM and handling of their output. Where such relations are deemed as relevant, the methods classes are aware of each other’s functionality and capable of responding appropriately to the output from other methods. This makes it possible to construct new routines by combining the available procedures, e.g. using a Log-likelihood Profiling within Case-deletion Diagnostics to assess the influence of individuals or groups on the confidence intervals of estimates of model parameters. PsN-Toolkit supports parallel execution on multiple processor computers as well as on clusters that support whole process migration such as openMosix.
Conclusions: Using PsN-Toolkit on computers with multiple processors or on computer clusters allows the application of methods in model building and validation that otherwise would have been practically impossible due to long run times.
References:
[1] B. Efron and R. Tibshirani, An introduction to the bootstrap, (Chapman & Hall, New York, 1993).
[2] J. S. U. Hjorth, Computer intensive statistical methods : validation model selection and bootstrap, (Chapman & Hall, London ; New York, 1994).
[3] L. Lindbom, R. Ribbing and E. N. Jonsson, Perl-speaks-NONMEM (PsN)––a Perl module for NONMEM related programming, Computer Methods and Programs in Biomedicine (In Press), Free download at http://psn.sourceforge.net.
[4] B. Szabó, Parallel::Forkmanager Perl Module, http://www.cpan.org/modules/by-module/Parallel/Parallel-ForkManager-0.7.5.tar.gz, (2/4 2003)
Reference: PAGE 13 (2004) Abstr 537 [www.page-meeting.org/?abstract=537]
Poster: poster