Physics-Informed Machine Learning for Personalized Disease Prediction, Prevention, and Management

This Collection supports and amplifies research related to SDG3, SDG9 and SDG10.

Physics-Informed Machine Learning (PI-ML) combines principles from physics- and biology-based modeling with data-driven Machine Learning (ML) methods. By embedding physical and biological laws—often expressed as differential equations—into ML frameworks, PI-ML provides models that require less experimental data, generalize better to unseen scenarios, and retain a degree of transparency, which are crucial for clinical trust and adoption.

This hybrid approach, blending data-driven learning with domain knowledge, addresses key limitations of both differential equations and black-box ML and combines the reliability of physics and biology with the flexibility of ML, making it particularly suited for biomedical and clinical applications.

In the context of personalized disease prediction, prevention, and management, PI-ML is particularly promising. For example, it can be applied to simulate tumor growth, cardiovascular flow, metabolic disorders, or infectious disease spread, among many other biomedical systems. By integrating mechanistic knowledge with patient-specific data, PI-ML facilitates personalized modeling and forecasting while maintaining scientific rigor. Furthermore, it enables the discovery of hidden dynamics and surrogate modeling for fast, accurate simulation in time-sensitive clinical settings.

This Collection welcomes original research, reviews, and perspectives focused on the use of PI-ML for modeling, predicting, and understanding disease processes to support personalized prevention and management.

 Topics of interest include, but are not limited to:

• Integration of differential equations-based models or multi-scale models into ML frameworks to develop approaches for disease modeling
• Solving inverse problems and uncovering hidden physical mechanisms in biomedicine using PI-ML to answer clinically relevant questions
• Surrogate modeling for real-time simulations and clinical decision support
• Physics-informed generative AI in healthcare
• Creation of digital twins and virtual twins for personalized healthcare using PI-ML techniques
• Modeling biomedical time-series, signals, and images using physics-informed methods in real-world settings
• Applications of PI-ML across different medical fields including, but not limited to oncology, neurology, cardiology, immunology, and endocrinology.
• Enhancing interpretability, explainability, and trustworthiness of PI-ML models for biomedical and clinical use

We encourage interdisciplinary submissions at the intersection of physics, biology, computational modeling, ML, and healthcare. Contributions should address methodological advances, theoretical insights, or impactful applications demonstrating the benefits of PI-ML in understanding, predicting, or managing disease, demonstrating potential for digital medicine innovation and clinical impact. Contributions are expected to demonstrate novelty, soundness of the proposed approach, potential for digital medicine innovation, and clinical relevance. Through this unique blend of disciplines and topics, this Collection aims to demonstrate how the emerging field of PI-ML can help address the complexity of disease modeling by concurrently enabling personalization and supporting computationally efficient assessment of high-dimensional problems, ultimately supporting the achievement of clinical impact.