John Metzcar,Catherine R. Jutzeler,Paul Macklin,Alvaro Köhn-Luque,Sarah C. Brüningk
Abstract:Mechanistic learning, the synergistic combination of knowledge-driven and data-driven modeling, is an emerging field. In particular, in mathematical oncology, the application of mathematical modeling to cancer biology and oncology, the use of mechanistic learning is growing. This review aims to capture the current state of the field and provide a perspective on how mechanistic learning may further progress in mathematical oncology. We highlight the synergistic potential of knowledge-driven mechanistic mathematical modeling and data-driven modeling, such as machine and deep learning. We point out similarities and differences regarding model complexity, data requirements, outputs generated, and interpretability of the algorithms and their results. Then, organizing combinations of knowledge- and data-driven modeling into four categories (sequential, parallel, intrinsic, and extrinsic mechanistic learning), we summarize a variety of approaches at the interface between purely data- and knowledge-driven models. Using examples predominantly from oncology, we discuss a range of techniques including physics-informed neural networks, surrogate model learning, and digital twins. We see that mechanistic learning, with its intentional leveraging of the strengths of both knowledge and data-driven modeling, can greatly impact the complex problems of oncology. Given the increasing ubiquity and impact of machine learning, it is critical to incorporate it into the study of mathematical oncology with mechanistic learning providing a path to that end. As the field of mechanistic learning advances, we aim for this review and proposed categorization framework to foster additional collaboration between the data- and knowledge-driven modeling fields. Further collaboration will help address difficult issues in oncology such as limited data availability, requirements of model transparency, and complex input data
What problem does this paper attempt to address?
The problem that this paper attempts to solve is in mathematical oncology, how to combine the advantages of knowledge - driven and data - driven modeling through mechanistic learning to more effectively promote the research of cancer biology and treatment. Specifically, the paper explores the following points:
1. **Definition and Background of Mechanistic Learning**:
- The paper first introduces the concept of mechanistic learning, that is, in mathematical oncology, a method that combines knowledge - based modeling (such as mechanistic mathematical models) with data - based modeling (such as machine learning and deep learning).
- It emphasizes the potential of mechanistic learning in dealing with complex cancer problems, especially in the case of limited data.
2. **Comparison between Knowledge - Driven and Data - Driven Modeling**:
- The paper compares in detail the advantages and disadvantages of knowledge - driven and data - driven modeling, including aspects such as model complexity, data requirements, output generation, and algorithm interpretability.
- It points out the differences between the two in terms of model transparency, data volume requirements, prediction ability, etc., and discusses how to overcome their respective limitations by combining their advantages.
3. **Four Combination Methods of Mechanistic Learning**:
- **Sequential Combination**: Knowledge - driven and data - driven modeling are applied sequentially, and the output of the former is used as the input of the latter.
- **Parallel Combination**: Knowledge - driven and data - driven modeling are carried out in parallel and complement each other.
- **External Combination**: Combine knowledge - driven and data - driven modeling at a higher level, such as digital twin technology.
- **Internal Combination**: Embed biomedical knowledge into the architecture or training process of data - driven models, such as Physics - informed Neural Networks (PINNs).
4. **Specific Application Cases**:
- The paper lists multiple application examples of mechanistic learning in oncology, including using neural networks as surrogate models, Neural Ordinary Differential Equations (Neural ODEs), digital twin technology, etc.
- Through these cases, it shows how mechanistic learning can improve the prediction ability and interpretability of models in actual research.
5. **Future Prospects**:
- The paper finally discusses the future development directions of mechanistic learning in mathematical oncology, emphasizing the importance of further cooperation to solve problems such as data availability, model transparency, and complex input data.
In summary, this paper aims to promote the development of mathematical oncology by integrating knowledge - driven and data - driven methods, and provide more reliable, data - supported, and interpretable models for cancer research and clinical treatment.