Learning Model Parameter Dynamics in a Combination Therapy for Bladder Cancer from Sparse Biological Data

Area of Science:

• Mathematical Biology
• Computational Oncology
• Machine Learning in Medicine

Background:

• Traditional models struggle with dynamic biological systems and sparse cancer data.
• Evolving interactions between tumor cells and immune cells are crucial for treatment response.
• Limited experimental data in oncology hinders accurate predictive modeling.

Purpose of the Study:

• To develop a novel framework for learning time-varying cell interactions in cancer.
• To predict subpopulation dynamics under anticancer treatments using limited data.
• To apply physics-informed neural networks (PINNs) to model evolving biological systems.

Main Methods:

• Utilized a physics-informed neural network (PINN) approach.
• Modeled interactions between bladder cancer cells and immune cells.
• Handled sparse, time-point-limited experimental data.

Main Results:

• Successfully predicted subpopulation trajectories at unobserved time points.
• Demonstrated consistency between model predictions and biological explanations.
• Showcased the framework's ability to capture evolving dynamics.

Conclusions:

• The proposed PINN method effectively models time-varying biological interactions.
• This approach offers a robust framework for analyzing cancer treatment dynamics with limited data.
• Enables prediction of cell behavior in response to interventions.