From shallow to deep: The evolution of machine learning and mechanistic model integration in cancer research

This review explores the integration of machine learning (ML) and mechanistic modelling to address challenges in computational biology, particularly in cancer research. While ML excels in processing large datasets and identifying complex, nonlinear relationships, mechanistic models provide causal insights grounded in biological principles. We classify the integration into shallow and deep categories. Shallow integration methods—such as sensitivity analysis, surrogate modelling, and data augmentation—have demonstrated improved computational efficiency and prediction accuracy. Deep integration goes further by embedding biological mechanisms directly into ML models, enhancing both explainability and performance in biological systems. Applications across cancer signalling, pharmacokinetics, and cell signalling illustrate the effectiveness of these integrated strategies. However, challenges including computational scalability and data quality must be addressed to fully realize their potential. We highlight key advancements in the integration of ML and mechanistic models and suggest that their continued evolution will drive future innovations in computational biology and systems medicine. • Integrating machine learning with mechanistic models enhances prediction accuracy and model interpretability. • Shallow integration methods, including surrogate modelling and data augmentation, improve computational efficiency. • Deep integration embeds biological mechanisms into machine learning models, advancing explainability in complex systems. • These hybrid methods have broad applications, from cancer to pharmacokinetics, with potential to improve precision medicine.