Finch Studio: A Next Generation NONMEM Modeling Workbench
Mohamed Ismail (1), Scott Van Wart (1), Donald E. Mager (1, 2)
(1) Enhanced Pharmacodynamics, LLC, (2) Department of Pharmaceutical Sciences, University at Buffalo, SUNY
Introduction: Finch Studio is an integrated modeling environment for pharmacometricians, clinical pharmacologists, and other team members to visualize, develop, and organize PK/PD models and data. The goal of Finch Studio is to increase modeler efficiency, lower the bar to entry and increase productivity for new pharmacometricians, and make PK/PD results more interpretable to non-modeler team members. It includes a powerful NONMEM code editor and workbench, a library of > 100 PK/PD models, and interactive data visualizations to help guide decision-making during the analysis. It has been designed to integrate nicely with existing tools such as Perl Speaks NONMEM (PsN) [1], xpose [2], and R.
Finch Studio will be released under a free academic license for students and instructors to further the training in the field of pharmacometrics, while a commercial license for organizations will also be available.
Workbench and Model Notebook
The Finch Studio modeling workbench was designed to aid the modeler in: writing and debugging NONMEM control streams, assessing model performance and viewing initial and final parameter estimates, iterating through the model development process, and extending the application to include interactive data analytic tools and custom run and post-processing scripts. Features include:
- An advanced NONMEM code editor with syntax highlighting, error checking, code completion, documentation on hover, and code folding.
- Ability to add, delete, edit, and duplicate (with final estimates) models, including as batch actions.
- Model descriptions, notes, tags, and color for model tracking and filtering.
- Display of key modeling results for all models (e.g., final objective function, significant digits, minimization status, etc.).
- Display of initial and final parameters (with the ability to transform between regular and log domain).
- Side-by-side comparison of results across models.
- View code differences between models to aid in quality assurance.
- Exporting of models and datasets to regulatory compliant file directory structures.
- Built-in system console.
- Automatic detection of PsN, NONMEM, and R path locations.
- Automatic creation of xpose formatted $TABLE records based on dataset covariates and model parameters.
Interactive Visualizations
Interactive visualizations were built into Finch Studio to allow the modeler to quickly explore their data, view diagnostic plots of model results, and make decisions about next steps in the modeling process. The plots are seamlessly available without the need of any external applications. Currently, the following plot types are available:
- Plots of the observed data
- Population observed data plots
- Individual observed data plots
- Covariate distribution and correlation plots
- Diagnostic plots
- Individual fits
- Residual diagnostic plots (CWRES vs Time, CWRES vs PRED, DV vs PRED, DV vs IPRED)
- Post-hoc parameter distribution and correlation plots
- Post-hoc parameters vs covariates
For all plot types, data can be filtered, colored, and stratified by covariates and study specific characteristics to quickly identify meaningful covariates and differences in subpopulations. Plots can be quickly exported as .png images.
Model Library
Finch Studio currently has a library of > 100 different combinations of possible initial modeling templates. To generate an initial model template, different model features can be selected including dosing route, number of compartments, absorption model, clearance mechanism(s), parameterization (e.g., rate constants vs. semi-physiological terms, built-in ADVANS vs. differential equations), and statistical models for parameters and residual unexplained variability.
Results
Finch Studio has been used internally since August 2021, as well as externally as part of a closed beta release since February 2022 by industry scientists and University at Buffalo students for the purposes of use in didactic coursework and for collecting feedback to enhance the application. The software has aided in the standardization of the modeling workflow, improving modeler efficiency (most notably for students and beginner modelers), and guiding collaborative, cross-functional discussions of modeling results and next steps. General release is planned for the Fall of 2022.
References:
[1] Karlsson MO, Nordgren R et al., PsN: An open source toolkit for non-linear mixed effects modelling, 2004-, https://uupharmacometrics.github.io/PsN/
[2] Jonsson EN, Karlsson MO. Xpose--an S-PLUS based population pharmacokinetic/pharmacodynamic model building aid for NONMEM. Computer Methods and Programs in Biomedicine. 58(1):51-64