Machine learning-based clustering identifies obesity subgroups with differential multi-omics profiles and metabolic patterns

Objective

Individuals living with obesity are differentially susceptible to cardiometabolic diseases. We hypothesized that an integrative multi-omics approach might improve identification of subgroups of individuals with obesity who have distinct cardiometabolic disease patterns.

Methods

We performed machine learning-based, integrative unsupervised clustering to identify proteomics- and metabolomics-defined subpopulations of individuals living with obesity (BMI ≥ 30 kg/m²), leveraging data from 243 individuals in the Multi-Ethnic Study of Atherosclerosis (MESA) cohort. Omics that contributed to the observed clusters were functionally characterized. We performed multivariate regression to assess whether the individuals in each cluster demonstrated differential patterns of cardiometabolic traits.

Results

We identified two distinct clusters (iCluster1 and 2). iCluster2 had significantly higher average BMI values, fasting blood glucose, and inflammation. iCluster1 was associated with higher levels of total cholesterol and high-density lipoprotein cholesterol. Pathways mediating cell growth, lipogenesis, and energy expenditures were positively associated with iCluster1. Inflammatory response and insulin resistance pathways were positively associated with iCluster2.

Conclusions

Although the two identified clusters may represent progressive obesity-related pathologic processes measured at different stages, other mechanisms in combination could also underpin the identified clusters given no significant age difference between the comparative groups. For instance, clusters may reflect differences in dietary/behavioral patterns or differential rates of metabolic damage.