Summary
A deep-learning model trained on 3D movement data from 3,328 adults significantly predicted 1,887 of 3,210 health-related targets. The Nature Medicine briefing reports predictive gains across all 18 body systems in men and 17 of 18 in women.
Researchers have developed a deep-learning model that uses 3D body movement to predict a broad range of health-related traits. A Nature Medicine Research Briefing published on 9 November 2026 reports that the model significantly predicted 1,887 of 3,210 phenotypic targets, with predictive gains across all 18 body systems in men and 17 of 18 in women.
How the gait model was built
The model was trained using skeletal motion data from 3,328 deeply phenotyped adults. A depth camera recorded their movement during five motor tasks. The resulting 3D data represent body joints and how they move over time.
The system produces numerical representations, known as embeddings, from the movement data. These representations can then be used to predict phenotypes: measurable characteristics related to health. The briefing also reports improved prediction of clinical diagnoses and medication use. Its figure caption says the model outperformed engineered gait features, which are measurements selected and calculated by researchers; the briefing does not give a numerical comparison.
What the findings could mean
Walking involves coordinated movement across the body, so gait data may contain signals associated with more than mobility or a single condition. The breadth of predictions reported here suggests that movement analysis could be useful for studying relationships between physical function and health across multiple systems.
The findings describe prediction from a research dataset, rather than a validated diagnostic process for individual patients. The briefing does not provide performance metrics or details of the validation design, which are important for judging how well the model might generalise beyond the study cohort.