Breath molecular feature profiling with machine learning for noninvasive diagnosis of gout: a single-center case-control study
- 1School of Clinical Medicine, Jiangxi University of Chinese Medicine , , ,
- 2Department of Orthopedics, Affiliated Hospital of Jiangxi University of Chinese Medicine , , ,
- 3Scientific Research Department, Affiliated Hospital of Jiangxi University of Chinese Medicine , , ,
Abstract
Objective: Gout, a purine metabolism disorder with growing global prevalence, is currently diagnosed by ACR/EULAR criteria, yet complementary noninvasive tools remain limited. We evaluated whether breath molecular feature analysis could discriminate established gout patients from healthy controls.
Methods: In this preliminary single-center study, breath samples were collected from 201 participants (99 established gout patients and 102 healthy controls). Breath molecular features were profiled by electrospray extraction ionization mass spectrometry (EESI-MS). Random forest machine-learning was applied to build classification models and identify discriminatory breath molecular features. Tongue-coating features and clinical characteristic were also compared between groups.
Results: EESI-MS profiling of exhaled breath identified 12 differential breath molecular features as candidate gout biomarkers. The putative urate dimer was significantly upregulated, whereas the other 11 ions were downregulated. The BreathGOUT model based on breath molecular features achieved an AUC of 0.999 (95% CI 0.997–1.000) by 5-fold stratified cross-validation, with accuracy, sensitivity, and specificity all reaching 0.990. Unimodal decomposition analysis confirmed that BreathGOUT retained an AUC of 0.999, indicating that breath molecular features provided discriminatory information complementary to that captured by clinical characteristics alone. The trimodal integrated model, BreathGOUT-Plus, also yielded an AUC of 0.999; although it did not further improve diagnostic performance, it enhanced robustness against short-term confounding influences.
Conclusion: In this case-control setting, machine-learning analysis of breath molecular features reliably discriminated established gout from healthy states. The model demonstrated excellent discriminative power and identified candidate biomarkers. Multimodal fusion improved clinical robustness and interpretability across Western and traditional Chinese medicine dimensions. This proof-of-concept supports future multicenter studies incorporating asymptomatic hyperuricemia, first-presentation arthritis, and other arthritides to assess broader clinical applicability.
Related
- Early diagnosis and molecular classification of gastric cancer peritoneal metastasis by deep learning-driven single-cell lipidomics
- Unmasking Early Hepatorenal Fibrosis in Polycystic Kidney Disease Using Noninvasive Precision Molecular MRI of Collagen
- Microfluidic Profiling of an Immune Cell‐Associated Extracellular Vesicle Signature for Active Tuberculosis Diagnosis and Treatment Monitoring
- Recent advances in biosensing platforms utilizing exosomal biomarker profiling for cancer diagnosis
- From conservation scores to language models: a feature engineering perspective on machine learning for genetic variant interpretation
BibTeX
@article{cai2026breath,
title = {{Breath molecular feature profiling with machine learning for noninvasive diagnosis of gout: a single-center case-control study}},
author = {Zebin Cai and Xin Huang and Xuao Luo and Xiaoling Chen and Weikang Sun and Hua-nan Li},
journal = {Frontiers in molecular biosciences},
year = {2026},
month = oct,
doi = {10.3389/fmolb.2026.1939939},
url = {https://europepmc.org/article/PMC/PMC13730828},
note = {CC BY},
}
RIS
TY - JOUR
TI - Breath molecular feature profiling with machine learning for noninvasive diagnosis of gout: a single-center case-control study
AU - Zebin Cai
AU - Xin Huang
AU - Xuao Luo
AU - Xiaoling Chen
AU - Weikang Sun
AU - Hua-nan Li
PY - 2026
DA - 2026/10/21
JO - Frontiers in molecular biosciences
DO - 10.3389/fmolb.2026.1939939
UR - https://europepmc.org/article/PMC/PMC13730828
AB - Objective: Gout, a purine metabolism disorder with growing global prevalence, is currently diagnosed by ACR/EULAR criteria, yet complementary noninvasive tools remain limited. We evaluated whether breath molecular feature analysis could discriminate established gout patients from healthy controls. Methods: In this preliminary single-center study, breath samples were collected from 201 participants (99 established gout patients and 102 healthy controls). Breath molecular features were profiled by electrospray extraction ionization mass spectrometry (EESI-MS). Random forest machine-learning was applied to build classification models and identify discriminatory breath molecular features. Tongue-coating features and clinical characteristic were also compared between groups. Results: EESI-MS profiling of exhaled breath identified 12 differential breath molecular features as candidate gout biomarkers. The putative urate dimer was significantly upregulated, whereas the other 11 ions were downregulated. The BreathGOUT model based on breath molecular features achieved an AUC of 0.999 (95% CI 0.997–1.000) by 5-fold stratified cross-validation, with accuracy, sensitivity, and specificity all reaching 0.990. Unimodal decomposition analysis confirmed that BreathGOUT retained an AUC of 0.999, indicating that breath molecular features provided discriminatory information complementary to that captured by clinical characteristics alone. The trimodal integrated model, BreathGOUT-Plus, also yielded an AUC of 0.999; although it did not further improve diagnostic performance, it enhanced robustness against short-term confounding influences. Conclusion: In this case-control setting, machine-learning analysis of breath molecular features reliably discriminated established gout from healthy states. The model demonstrated excellent discriminative power and identified candidate biomarkers. Multimodal fusion improved clinical robustness and interpretability across Western and traditional Chinese medicine dimensions. This proof-of-concept supports future multicenter studies incorporating asymptomatic hyperuricemia, first-presentation arthritis, and other arthritides to assess broader clinical applicability.
SN - 2296-889X
ER -