Bridging mechanism and data: Hybrid modeling approaches for cancer and aging research

Artikel i vetenskaplig tidskrift, 2026

Systems biology continues to face the challenge of uniting causal explanation and interpretability with the drive for greater predictive power and scalability. Mechanistic models based on ordinary differential equations (ODEs) provide interpretability and causal grounding in systems biology, yet they often suffer from parameter uncertainty, limited scalability, and computational costs. Machine learning (ML) approaches offer strong predictive performance by learning from high-dimensional, noisy biological data, but this data-driven strength comes at the cost of limited transparency and limited generalizability. Hybrid approaches that integrate mechanistic modeling with ML are emerging as a powerful new paradigm: data-driven modules reduce dimensionality and noise, encode multimodal and longitudinal data, and serve as surrogates for expensive mechanistic submodels, while mechanistic constraints guide ML toward biologically meaningful solutions. This synergy opens the door to uncertainty-aware, generalizable, and computationally tractable models with enhanced predictive power. Applications in cancer and aging research illustrate the promise of hybrid models in predicting treatment success, charting aging trajectories, and designing preventive strategies. Hybrid mechanistic-ML frameworks are not merely incremental improvements but represent a step towards personalized digital twins of biological systems, adaptive, interpretable, and predictive tools for precision medicine and geroscience.