Deep learning approach to parameter optimization for physiological models.

Area of Science:

• Computational Biology
• Systems Biology
• Biophysics

Background:

• Inferring nonlinear dynamics and parameters in biological data modeling is challenging.
• Standard parameter optimization methods struggle with biological range constraints for nonlinear models.

Purpose of the Study:

• Propose a novel method using neural networks for biological modeling, parametrization, and parameter inference.
• Evaluate and improve putative models by simultaneously addressing these challenges.
• Adapt a deep learning framework for parameter inference in various physiological systems.

Main Methods:

• Utilized clinical frequently sampled intravenous glucose tolerance testing data.
• Introduced two physiological lipolysis models for glucose, insulin, and free fatty acids dynamics.
• Trained a convolutional neural network on simulated time course data for parameter inference.

Main Results:

• The trained neural network achieved accurate parameter inference and trajectory reconstruction across different settings.
• Consistently high R-squared values and low P-values were observed.
• Feature engineering and training dataset size impacted inference performance, with specific choices improving accuracy.

Conclusions:

• Established a deep learning framework for parameter inference in mathematical models.
• Demonstrated the framework's adaptability to various physiological systems.
• Showcased the potential for improved biological data modeling and parameter estimation.