Logo image
Physics-inspired neural network for a pseudo-digital twin model of P2X7 receptor signaling pathway dynamics
Thesis   Open access

Physics-inspired neural network for a pseudo-digital twin model of P2X7 receptor signaling pathway dynamics

Adesina Adewole
Master of Science (MSc), University of Johannesburg
2025
Handle:
https://hdl.handle.net/10210/520801

Abstract

Cancer progression is significantly affected by signaling cascades, with the Purinergic receptor (P2X7R) pathways playing a role in the regulation of the programmed cell death through Adenosine Triphosphate (ATP)-mediated activation. The heterogeneity of the P2X7R and downstream activation of proliferation across individuals, makes therapeutic prediction challenging. This research focus on modeling the P2X7 receptor pathway to explore how different kinetic parameters influence system behavior. We develop a computational model that is capable of simulating individual specific response, This model can be use for personalized drug response and precision diagnostics. The system was modeled using the fourth-order Runge-kutta (RK4) method to solve the ordinary differential equations(ODEs) that governs the P2X7 receptor dynamics. However, this approach encounter some limitations such as computational inefficiency and numerical instability which limit the progress of the biological systems. Also, developing a data driven model was a challenge due to lack of experimental data. To address this challenges, a physics-Inspired Neural Network (PINN) model was developed. This model combine the strength of ODE modeling and data-driven learning by incorporating the physical law of P2X7R signaling into the neural network training process. As a result of this, PINN enhance model robustness and reduces computational stiffness which adapt to variation in tumor micro-environmental parameters. In this form, the Physics Inpired Neural Network (PINN) model developed operates as pseudo-digital twin, a computational replica that interpret the individual-specific signaling behavior that based on periodic simulation data instead of a continuous real time data input. The trained PINN achieved a low mean absolute error (MAE < 0.027) across all simulated plots which demonstrate that the PINNs approximate the ODEs trajectories across different individual kinetic parameter and establish a foundation for individualspecific digital twins in cancer research and offering a pathways toward predictive tumor modeling
pdf
ADESINA2.57 MBDownloadView
Open Access

Metrics

1 Record Views

Details

Logo image