Gait foundation model links walking patterns to hundreds of traits in Nature Medicine

A peer-reviewed study in Nature Medicine (published 6 November 2026) describes a gait foundation model built from 3D skeletal motion captured with a single depth camera during five motor tasks in 3,328 deeply phenotyped adults from the Human Phenotype Project (Segal lab and colleagues).

Main findings

Learned embeddings outperformed engineered gait features on many targets. Correlations included age r=0.69, BMI r=0.91, and visceral adipose tissue area r=0.83. The model significantly predicted 1,887 of 3,210 tested phenotypes. Versus a baseline of age, BMI, VAT and height, prediction improved across all 18 physiological systems in males and 17 of 18 in females. Leg-related signals mapped more toward metabolic and frailty traits; torso-related signals toward sleep and lifestyle traits.

Limits — not a diagnostic tool

This is an observational, cross-sectional cohort study, not a clinical trial or diagnostic device. Competing interests include Newton VR employees and Pheno.AI consultants; data access is privacy-gated. Findings describe associations in this cohort and do not replace clinical assessment. This is not personal medical advice — talk to your doctor about any health concerns.