Systems biology informed deep learning for inferring parameters and hidden dynamics.

Area of Science:

• Systems Biology
• Computational Biology
• Biophysics

Background:

• Mathematical models of biological reactions generate complex systems of ordinary differential equations with numerous unknown parameters.
• Parameter inference and prediction of hidden dynamics from limited experimental data are critical challenges in systems biology.

Purpose of the Study:

• To develop a novel, robust algorithm for parameter inference and dynamic prediction in systems biology.
• To integrate ordinary differential equations directly into a deep learning framework to constrain the model.

Main Methods:

• Developed a systems-biology-informed deep learning algorithm.
• Incorporated ordinary differential equations into neural network architecture.
• Utilized few, scattered, and noisy experimental measurements for training and validation.

Main Results:

• Successfully inferred dynamics of unobserved species and external forcing.
• Accurately estimated unknown model parameters.
• Demonstrated algorithm efficacy on three different benchmark problems.

Conclusions:

• The developed deep learning algorithm effectively infers parameters and dynamics in biological systems.
• Integrating ordinary differential equations provides a powerful constraint for data-driven modeling.
• This approach offers a robust solution for analyzing complex biological data with limited measurements.