AI-Aristotle: A physics-informed framework for systems biology gray-box identification.

Area of Science:

• Systems Biology
• Computational Biology
• Biophysics

Background:

• Discovering mathematical equations for biological systems from data is a critical scientific challenge.
• Existing methods often struggle with parameter estimation and identifying unknown physical laws (gray-box identification).

Purpose of the Study:

• To present AI-Aristotle, a novel physics-informed framework for parameter estimation and gray-box identification in Systems Biology.
• To evaluate the framework's accuracy, speed, flexibility, and robustness using benchmark problems.

Main Methods:

• AI-Aristotle combines eXtreme Theory of Functional Connections (X-TFC) and Physics-Informed Neural Networks (PINNs) with symbolic regression (SR).
• The framework utilizes domain decomposition and integrates neural networks with symbolic regressors.
• Performance is assessed using sparse synthetic data with added noise.

Main Results:

• AI-Aristotle demonstrated accuracy and robustness in parameter estimation and gray-box identification on pharmacokinetic and glucose-insulin models.
• Comparisons were made between X-TFC and PINNs, with cross-verification using two distinct SR techniques.
• The study provides insights into the performance trade-offs of integrating neural networks and symbolic regression.

Conclusions:

• AI-Aristotle offers a comprehensive approach for gray-box identification in complex dynamical systems.
• The framework provides valuable guidance for researchers in biomedicine and other fields dealing with data-driven scientific discovery.
• Integrating neural networks with symbolic regression enhances the ability to uncover underlying physical principles from observational data.