In biomedical modeling, the integration of mechanistic and data-driven approaches is reshaping how we interpret and predict complex biological phenomena. Mechanistic or “white-box” models offer transparency and generalizability by representing biological processes via first-principles, while artificial intelligence excels at uncovering intricate patterns in high-dimensional data streams. The challenge arises in reconciling these methods to address limitations inherent in each—mechanistic models may be limited by incomplete or imprecise knowledge, whereas AI methods can fall short in interpretability and generalization beyond the training data. Recent advances have seen increasing use of “grey-box” models, which blend mechanistic structure with data-driven techniques, enabling both predictive accuracy and scientific insight. However, the field remains fragmented, with diverse strategies for merging model types, encoding biological knowledge, and managing uncertainties, highlighting an urgent need for systematic evaluation, methodological innovation, and consensus on best practices.

This Research Topic aims to catalyze methodological progress in grey-box modeling within biomedicine and the life sciences. Central objectives include clarifying effective strategies for hybridization, rigorously analyzing identifiability and validation schemes, and promoting approaches that enhance interpretability and trustworthiness. Specific questions driving this Research Topic include: What are the most reliable routes for integrating mechanistic knowledge with AI? How can hybrid models be validated for scientific and practical use? What frameworks best capture and quantify uncertainty in predictions, and how can these be communicated transparently for clinical or scientific deployment?

The scope of this Research Topic encompasses methods-driven research advancing the construction, evaluation, and deployment of hybrid mechanistic/AI models across the molecular, cellular, tissue, and organismal levels. Review articles, original research, benchmark studies, and conceptual analyses are welcomed, provided they yield clear methodological insights beyond strictly applied case studies. To gather further insights in this rapidly evolving field, we welcome articles addressing, but not limited to, the following themes:

• Development of mechanistic cores augmented with machine learning surrogates for computationally intensive modules or simulators
• Design and analysis of physics-/biology-informed ML and neural differential equations constrained by mechanistic principles
• Grey-box system identification and hybrid state-space modeling, including data-driven inference of unknown terms in dynamical models
• Machine learning-assisted parameter estimation, data assimilation, and solutions to inverse problems in biological systems
• Approaches for uncertainty quantification, model calibration, robustness, and reliability of hybrid predictions
• Techniques for enhancing identifiability, interpretability, and mechanistic plausibility in hybrid modeling architectures
• Multi-fidelity and multi-scale modeling workflows, including coupling with omics, imaging, or electronic health record data
• Establishment of benchmark datasets, validation protocols, and best practices for comparing integration strategies
• Methodologically innovative case studies demonstrating clear advances in hybrid mechanistic/AI modeling

We welcome original research articles, methodological papers, systematic reviews, conceptual/perspective articles, and benchmark datasets or protocols.

Authors

Frontiers in Systems Biology

Multiscale Mechanistic Modeling