Nature Reviews. Cancer | Applications of Mathematical Modeling and Artificial Intelligence in Cancer Therapy Delivery and Efficacy Optimization

This article systematically reviews recent advances in mathematical modeling and artificial intelligence for enhancing the delivery efficiency and therapeutic efficacy of cancer treatments, highlighting the synergistic potential between mechanistic models and AI approaches, thus providing new avenues for personalized cancer therapy.

 

Literature Overview

The article 'Using mathematical modelling and AI to improve delivery and efficacy of therapies in cancer', published in Nature Reviews. Cancer, reviews and summarizes the applications of mathematical modeling and artificial intelligence in predicting and optimizing the delivery and efficacy of molecular, antibody, nanotherapeutic, and cell-based therapies in solid tumors. The article systematically explains the role of mechanistic models—including lumped-parameter and distributed-parameter models—in describing drug pharmacokinetics, the influence of the tumor microenvironment, and responses to immunotherapy. It also explores the potential of AI methods in processing high-dimensional data and identifying treatment biomarkers, proposing the integration of both approaches as a future direction for enabling precision oncology.

Background Knowledge

Solid tumor therapy faces multiple physiological barriers, including abnormal tumor vasculature, dense extracellular matrix (ECM), elevated interstitial fluid pressure (IFP), and solid stress, all of which collectively limit effective drug delivery and penetration. Additionally, immunosuppressive microenvironments further reduce the efficacy of immunotherapies. In recent years, mathematical modeling has become an essential tool for understanding these complex biophysical processes. Mechanistic models, based on biological principles, use ordinary differential equations (lumped-parameter) or partial differential equations (distributed-parameter) to describe the spatiotemporal distribution of drugs in tissues, integrating drug physicochemical properties, tumor structural features, and dynamic changes in the microenvironment. Meanwhile, artificial intelligence—particularly machine learning—has demonstrated significant potential in cancer early detection, prognosis prediction, and drug discovery by uncovering hidden patterns in large-scale omics and imaging data. However, each method has limitations: mechanistic models rely on precise parameters, while AI models are often seen as 'black boxes.' Therefore, combining the interpretability of mechanistic models with the data-driven power of AI has become a key breakthrough for improving treatment prediction accuracy and designing personalized strategies. This study systematically reviews existing modeling frameworks and advocates for an integrated paradigm to advance cancer therapy optimization.

 

 

Research Methods and Experiments

This article adopts a review-based methodology to systematically summarize recent applications of mathematical modeling and artificial intelligence in the field of cancer therapy delivery and efficacy prediction. The study first categorizes mechanistic models, including lumped-parameter models (e.g., pharmacokinetic/PK-PD models, PBPK models) used to describe time-dependent concentration changes of drugs in the whole body or tumor tissues, and distributed-parameter models (e.g., partial differential equation models) used to simulate the spatiotemporal distribution of drugs within tumors and the effects of physical factors such as interstitial pressure and solid stress. Additionally, the article reviews the applications of discrete models (e.g., cell-based models) and stochastic models in simulating cellular behaviors, angiogenesis, and nanoparticle transport. Subsequently, the study summarizes artificial intelligence methods—including traditional machine learning (e.g., logistic regression, support vector machines, gradient boosting) and deep learning (e.g., neural networks, convolutional networks)—in processing medical images, omics data, and clinical information for predicting treatment responses, identifying biomarkers, and optimizing therapeutic regimens. The article also emphasizes synergistic strategies such as using simulation data generated by mechanistic models to train AI models, and employing AI for parameter identification and model simplification.

Key Conclusions and Perspectives

• Mechanistic mathematical models can effectively simulate drug delivery processes in tumors, revealing how physical barriers such as interstitial fluid pressure, solid stress, and vascular abnormalities limit therapeutic efficacy
• Lumped-parameter models are suitable for describing average drug concentration changes in the whole body or tissues, whereas distributed-parameter models can simulate heterogeneous drug distribution within tumors, supporting personalized treatment design
• Discrete and stochastic models can capture cellular-level heterogeneity and dynamic behaviors, offering a microscopic perspective on drug–cell interactions within the tumor microenvironment
• Artificial intelligence methods have advantages in processing high-dimensional data, enabling the identification of complex patterns and biomarkers associated with treatment response from imaging and omics data
• Integrating mechanistic models with AI allows the generation of large training datasets from simulations to enhance the generalization ability of AI models, while AI can accelerate parameter identification and simulation of complex models
• 'Digital twin' models that integrate medical imaging (e.g., MRI, CT, elastography) with multi-omics data hold promise for patient-specific treatment prediction and optimization

Research Significance and Prospects

This study systematically integrates recent advances in mathematical modeling and artificial intelligence for optimizing cancer therapy, emphasizing the importance of cross-scale modeling frameworks. By using mechanistic models to dissect biophysical processes and combining them with AI to extract deep features from data, it provides new insights into overcoming drug delivery barriers and improving the accuracy of treatment response prediction. Future research directions include developing more sophisticated multiscale integrated models, incorporating real-time clinical data for dynamic updates, and advancing the clinical application of such models in trial design and personalized treatment decision-making.

 

 

Conclusion

This article provides a comprehensive review of the synergistic roles of mathematical modeling and artificial intelligence in enhancing the delivery efficiency and therapeutic efficacy of cancer treatments. Mechanistic models offer interpretable predictive frameworks for drug distribution and treatment response by describing the physical and biological characteristics of the tumor microenvironment. At the same time, AI methods leverage powerful data-learning capabilities to identify potential therapeutic biomarkers and response patterns from complex clinical and omics data. Their integration not only overcomes the limitations of individual approaches but also advances the development of personalized treatment strategies such as 'digital twins.' In the future, integrating multimodal data and achieving clinical translation of these models will be key to realizing precision oncology. This study provides important theoretical support and technical guidance for developing more effective therapies, optimizing dosing strategies, and improving patient outcomes.

 

Constantinos Harkos, Andreas G Hadjigeorgiou, Chrysovalantis Voutouri, Triantafyllos Stylianopoulos, and Rakesh K Jain. Using mathematical modelling and AI to improve delivery and efficacy of therapies in cancer. Nature reviews. Cancer.