Online Program

Assessment of Inflammatory Biomarkers Related to Obesity in Adolescents Using K-Means Clustering and Random Forests Algorithms (309865)

Keywords: K-means clustering, Random Forests, inflammatory markers, obesity

It has been demonstrated that obesity is associated with low-grade inflammatory process that may have a casual rule in the development of cardiovascular diseases, metabolic syndrome, insulin resistance, and diabetes mellitus. We used two machine learning algorithms to determine the relationships between inflammatory markers ((IL-6, IL-8; TNF-a, IL- 1ß, NGF, MCP-1, HGF, CRP, and Leptin) and obesity in Mexican American adolescents. K-means and Hierarchical cluster (using Ward method) analyses were performed on log-transformed and standardized biomarker to identify adolescents' groups having similar biomarker patterns. clValid package in R was used to assess the quality and stability of clusters generated using k-means and other algorithms. Random Forest (RF) method was used to ascertain the importance of the effect of inflammation markers on BMI levels, waist circumference (WC), waist-to-hip ratio (WHR). We derived 3 K-means clusters: cluster 1 was characterized by high CRP and Leptin levels; cluster 2 had elevated IL-8, TNF-a, MCP-1, and HGF levels; and cluster 3 had the lowest levels of all inflammatory markers. Individuals in cluster 1 had significantly higher mean BMI levels, WC, and WHR (Tukey-Kramer adjusted p-values <0.0001) compared to clusters 2 and 3. Based on a RF with 500 trees, Leptin and CRP had the higher important effect on BMI, WC, and WHR.