Gait embeddings from a single depth camera predict health phenotypes across 18 body systems
A gait foundation model that turns three-dimensional skeletal motion, recorded by a single depth camera during five motor tasks, into learned embeddings predicted age (), body mass index () and visceral fat area () better than engineered gait features in 3,328 deeply phenotyped adults, significantly predicted 1,887 of 3,210 tested phenotypes, and improved on a baseline of age, body mass index, visceral fat and height in all 18 body systems in men and 17 of 18 in women, including clinical diagnoses and medication use.
- Model
- Gait foundation model producing learned embeddings of 3D skeletal motion; architecture not named in the abstract
- Field
- Medicine
- Date
- 2026-11-06
- Human collaborators
- Adam Gabet, Sarah Kohn, Guy Lutsker, Adina Weinberger, Eran Segal
What was found
Gait is usually summarised by engineered measurements such as cadence and stride length and read as a sign of particular diseases. Here the model encodes how all body joints move together over time, and those embeddings were tested against thousands of phenotypes measured in the Human Phenotype Project cohort, from grip strength to sleep quality. Ablations that removed body regions found leg dynamics carried most of the metabolic and frailty signal and torso dynamics most of the sleep and lifestyle signal. The authors argue for gait as a passive, multisystem vital sign in routine clinical assessment.
Novelty check
Europe PMC, 2026-11-12: gait in the title with foundation model, deep learning, embedding or machine learning, and health, phenotype, disease or biomarker, 78 records. Nearly all prior work predicts one condition, most often Parkinson's disease, falls or multiple sclerosis, from gait; the only multisystem phenome-wide records are this paper and its own research briefing. No earlier entry in this registry covers gait.
Caveats and known objections
Peer-reviewed in Nature Medicine, by the developers, with no independent replication; the paper itself was not read in full here. The machine-learning element is the learned embedding; most of the finding is an association analysis showing that gait carries information about many body systems, so the AI contribution is a representation that humans then tested. The abstract describes one Israeli cohort of 3,328 adults and reports no external cohort, and its outcomes are phenotypes measured within that cohort, many of them surrogates, rather than future clinical events. Competing interests per the paper: three authors are employees of Newton VR Ltd., which developed the Movement Lab system used to record gait, and two senior authors are paid consultants of Pheno.AI, Ltd. Autonomy is ai-assisted: humans designed the study and the analyses, and the model supplied the features.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Peer reviewed and AI-assisted.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded peer reviewed for verification and ai-assisted for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). Gait embeddings from a single depth camera predict health phenotypes across 18 body systems. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-11-06-gait-foundation-model-phenotypes
BibTeX
@misc{whataifound-weizmanninstituteofscience-2026-phenotypes,
title = {Gait embeddings from a single depth camera predict health phenotypes across 18 body systems},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by Weizmann Institute of Science. Verification: Peer reviewed. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2026-11-06-gait-foundation-model-phenotypes}
}