Related Experiment Video
Updated: Aug 19, 2026
In Vitro Modeling of Fat Deposition in Metabolic Dysfunction-Associated Steatotic Liver Disease
Published on: July 19, 2024
Association between UMAP-identified body composition phenotypes and metabolic dysfunction-associated steatotic liver
Huimin Yin1, Yangtian Wang2, Li Yuan2
1Department of Clinical Nutrition, Taikang Xianlin Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, China.
Background:
Metabolic dysfunction-associated steatotic liver disease (MASLD) poses a critical global health challenge, yet traditional anthropometric indices fail to precisely capture its metabolic heterogeneity. This study aimed to identify novel body composition phenotypes using nonlinear dimensionality reduction and to develop a predictive model for MASLD.
Methods:
A total of 7,946 participants were enrolled. Uniform Manifold Approximation and Projection (UMAP) was applied to bioelectrical impedance analysis (BIA) data for unsupervised clustering to identify latent phenotypes. Key variables were selected using a combination of Boruta, XGBoost, and Random Forest algorithms, followed by the construction of a Logistic regression model to evaluate associations and predictive performance.
Results:
UMAP robustly identified three distinct phenotypes: Low-Mass Lean (LML), High-Mass Obesogenic (HMO), and Fluid-Imbalanced, High-Parameters (FIHP). In terms of frequency, LML was the most common (47.7% in males, 49.2% in females), followed by HMO (45.8% and 44.8%, respectively) and the rare FIHP subtype (6.5% and 6.0%, respectively). LML exhibited normal size and metabolic balance, whereas HMO was characterized by high adiposity and metabolic risk, and FIHP by fluid imbalance. Compared to the LML group, HMO and FIHP phenotypes were associated with significantly higher risks of MASLD (adjusted OR: 3.10 [95% CI: 2.74-3.51] and 1.39 [95% CI: 1.08-1.78], respectively). The integrated model demonstrated superior predictive performance (AUC = 0.863) and favorable clinical utility in calibration and decision curve analyses.
Conclusion:
Our study identified three distinct body composition phenotypes significantly associated with MASLD risk using unsupervised clustering and demonstrated that a prediction model integrating these phenotypes with routine indicators exhibited superior performance, suggesting their potential utility for early risk stratification and clinical management.
