Model predicts 1,887 health phenotypes from 3D movement data of 3,328 adults

Researchers evaluated three-dimensional skeletal motion from 3,328 adults to predict health biomarkers across 18 body systems, according to a study published on nature.com nature.comnature.comnature.comOpen source ↗. Using a single-depth camera, movement was recorded across five standardized motor tasks to capture whole-body coordination over time. The generated motion embeddings successfully predicted 1,887 of 3,210 evaluated clinical phenotypes.

Multisystem Biosignals

The motion model translates joint movement into learned numerical representations rather than standard metrics like stride length or cadence. These embeddings outperformed engineered gait features in estimating age (Pearson's r = 0.69), body mass index (r = 0.91), and visceral adipose tissue area (r = 0.83). When evaluated alongside baseline variables including age, height, body mass index, and visceral fat, the gait representations provided additive predictive value across all 18 systems in males and 17 of 18 in females.

Anatomical Dynamics

Ablation analyses indicated that different body segments correspond to distinct physiological profiles. Lower-limb kinematics served as the primary predictor for metabolic phenotypes and frailty markers. In contrast, torso motion dynamics showed the strongest associations with sleep metrics and lifestyle indicators. The findings indicate that continuous motion tracking can register physiological status across systemic domains without invasive diagnostic tools.