Modeling tumor progression and immune response is a key challenge in computational oncology. Traditional ODE-based models offer insights into interactions among cancerous, healthy, and immune cells, but often rely on ideal assumptions and dense data. In this research we explored the use of Physics-Informed Neural Networks (PINNs) to learn and predict the behavior of a nonlinear system modeling tumor-immune-normal dynamics. By embedding biological equations into the learning process, PINNs can train on sparse or noisy data while respecting domain constraints. We assess their performance under varying data availability, showing that moderate training data enables accurate reconstruction and extrapolation, whereas excessive data may induce localized errors. These findings suggest that PINNs are promising tools for biomedical modeling, with potential applications in personalized simulation and treatment planning.

This research will be presented in International Conference of the Chilean Computer Science Society (SCCC)