AI powered multi scale data integration paves the way for precision exercise biomedicine

A recent comprehensive review highlights that artificial intelligence, by integrating multiscale data from wearables, multiomics, imaging and largescale cohorts, can finally turn the “exercise is medicine” slogan into truly individualized clinical practice. This openaccess article entitled “Artificial intelligenceempowered multiscale data integration for precision exercise biomedicine” (DOI: 10.1515/teb20260009), was published in Translational Exercise Biomedicine (ISSN: 2942-6812), an official partner journal of International Federation of Sports Medicine (FIMS).

Exercise is widely recognized as a cornerstone for preventing and managing chronic diseases, including cardiovascular, metabolic, musculoskeletal and neurological disorders. Yet despite its wellestablished benefits, onesizefitsall exercise prescriptions often fail because individual physiological and molecular responses vary substantially. Differences in genetic background, baseline fitness, age, sex, comorbidities and environmental context all contribute to this heterogeneity, limiting the translation of exercise science into personalized clinical care.

The review identifies a rapidly expanding data ecosystem that is beginning to capture this complexity. Wearable technologies continuously record heartrate dynamics, activity patterns and sleep. Multiomics profiling provides molecularresolution readouts of exerciseinduced adaptation across the transcriptome, proteome and metabolome. Largescale biobanks and longitudinal cohorts link physical activity exposures to clinical outcomes, while imaging biomarkers offer structural and functional context. AI has matured into a powerful computational tool that can learn predictive representations from such heterogeneous data. 

Professor Ge Zhilei, Associate Professor at the School of Chemistry and Chemical Engineering of Shanghai Jiao Tong University and Vice Dean of the Institute of Translational Medicine, commented on the significance of the work: “This review systematically outlines how AI can bridge the gap between the rich multiscale data generated by modern exercise biomedicine and the urgent clinical need for truly personalized exercise prescriptions. By integrating wearables, multiomics and imaging, AI has the potential to transform exercise from a general recommendation into a precisely tailored, evidencebased intervention.”

The authors systematically examine four key AI modeling paradigms. Timeseries learning captures temporal dependencies in highfrequency physiological signals, using architectures such as long shortterm memory (LSTM) networks and transformers. Multimodal fusion integrates complementary data types, for example, combining wearable streams with imaging, omics and clinical variables, to exploit crossmodal complementarities. Causal inference methods, including instrumental variables and Mendelian randomization, help establish whether an exercise intervention causes an outcome change when randomized trials are infeasible. Reinforcement learning optimizes sequential decisions under uncertainty, offering a pathway to adaptive, safe exercise prescriptions that can be updated as patient data accumulate.

Emerging clinical applications are already taking shape. The review highlights work in cardiometabolic disease, metabolic liver disease, neurodegeneration, cancer survivorship and chronic kidney disease. In type 2 diabetes, for instance, reinforcement learning has been explored for personalized glycemic control, while causal inference frameworks have been used to link sustained physical activity trajectories to ageing outcomes. Foundation models trained on largescale unlabeled datasets, already successful in retinal imaging and pathology, suggest a future direction in which unified AI systems could simultaneously analyze wearable time series, molecular profiles and imaging data to generate personalized exercise recommendations.

However, the authors also identify persistent barriers to translation. Data heterogeneity, reliability under distribution shift, interpretability and uncertainty communication, as well as regulatory and ethical constraints, all need to be addressed before AIguided exercise prescriptions can enter routine clinical use. This review stresses that progress will depend on frameworks that link mechanistic insight with applicable digital monitoring and safetyaware adaptive prescriptions.

Professor Song Haitao, Dean of the Shanghai Artificial Intelligence Research Institute affiliated with Shanghai Jiao Tong University and IET Fellow, added: “The integration of AI into exercise biomedicine represents a paradigm shift. By harnessing advanced machine learning algorithms, we can now move beyond populationlevel guidelines and deliver individualized, adaptive exercise regimens that maximize therapeutic benefits while minimizing risks, an approach that could redefine preventive and rehabilitative medicine.”

The review serves as a milestone in the emerging field of precision exercise biomedicine, laying out both the opportunities and the challenges ahead. As wearable devices become ubiquitous and biobankscale datasets accumulate, AIdriven multiscale data integration may soon turn exercise prescriptions into truly personalized therapeutics.