pyDarwin: A Machine Learning Enhanced Automated Nonlinear Mixed-Effect Model Selection Toolbox.

Area of Science:

• Pharmacometrics
• Computational Biology
• Machine Learning

Background:

• Nonlinear mixed-effect (NLME) models are crucial in pharmacometrics for analyzing complex biological data.
• Traditional stepwise model selection methods can be labor-intensive, subjective, and may compromise model interpretability.
• There is a need for efficient and objective automated model selection tools in pharmacometrics.

Purpose of the Study:

• To introduce pyDarwin, an open-source Python package designed for automated nonlinear mixed-effect model selection.
• To demonstrate the pyDarwin workflow for global model search using machine learning and NONMEM.
• To provide a tutorial for researchers to efficiently perform robust and interpretable model selection.

Main Methods:

• pyDarwin integrates machine learning algorithms with the NONMEM software for global model space exploration.
• The package facilitates an objective and less labor-intensive model selection process.
• A user-defined model search space is explored to identify the optimal model.

Main Results:

• pyDarwin enables an efficient, objective, and robust model selection process.
• The package maintains model interpretability, a key advantage over traditional methods.
• The tutorial demonstrates the practical application of pyDarwin using real-world clinical data.

Conclusions:

• pyDarwin offers a powerful and accessible tool for advancing nonlinear mixed-effect model selection in pharmacometrics.
• The open-source nature and efficient workflow of pyDarwin promote wider adoption and reproducibility.
• This approach streamlines the discovery of optimal, interpretable models from complex datasets.