A Biology-Informed Deep Learning Framework for Continuous Modeling of Cancer Immunotherapy

Author

Meet Gupta, Amey Rathore, Adarsh Kumar, Akshat Singh

Abstract

 Traditional mathematical modeling of tumorimmune dynamics relies on Ordinary Differential Equations (ODEs) that struggle to adapt to patient-specific data, while standard deep learning approaches require vast datasets rarely available in clinical oncology. In this paper, we propose a BiologyInformed Neural Network (BINN) that embeds Lotka-Volterrastyle cellular interaction dynamics directly into the neural network’s loss function. By utilizing automatic differentiation, the model acts as a continuous function approximator that enforces strict biological constraints—specifically tumor growth, immune clearance, and simulated immunotherapy injections—without requiring large longitudinal datasets. Our experiments demonstrate that the BINN successfully models natural tumor escape. Furthermore, by introducing a mathematically smooth immunotherapy injection profile, the network accurately predicts the threshold required for effector T-cells to drive the tumor population to eradication. We also address network alignment issues, demonstrating how applying a Softplus activation successfully prevents non-physical cellular population predictions. Ultimately, this hybrid mathematical-machine learning framework provides a robust, data-efficient method for simulating continuous cancer trajectories, offering a powerful new tool for personalized immunotherapy dosage prediction.

Keywords

Biology-Informed Neural Networks, TumorImmune Dynamics, Physics-Informed Machine Learning, Mathematical Oncology, Immunotherapy.

Full Text:

Download Paper PDF

References

• Kirschner and J. C. Panetta, “Modeling immunotherapy of the tumorimmune interaction,” Journal of Mathematical Biology, vol. 37, no. 3, pp. 235–252, 1998.
• G. de Pillis, W. Gu, and A. E. Radunskaya, “Mathematical model of solid tumor growth and immunotherapies,” Journal of Theoretical Biology, vol. 238, no. 4, pp. 841–862, 2006.
• Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” Journal of Computational Physics, vol. 378, pp. 686–707, 2019.
• A. Itza, et al., “Using Physics-Informed Neural Networks (PINNs) for Tumor Cell Growth Modeling,” Mathematics, vol. 12, no. 8, 1195, 2024.
• Paszke et al., “PyTorch: An Imperative Style, High-Performance Deep Learning Library,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 32, 2019, pp. 8024–8035.
• P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” 3rd International Conference on Learning Representations (ICLR), 2015.
• Dugas et al., “Incorporating Second-Order Functional Knowledge for Better Option Pricing,” in Advances in Neural Information Processing Systems (NIPS), vol. 13, 2001, pp. 472–478.