UMAP-Based Visualization of PCB Congener Fingerprints Across Demographic Groups in the U.S. Population

Résumé fourni par la source

This study applies dimensionality reduction techniques, including principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), and uniform manifold approximation and projection (UMAP), to high-dimensional polychlorinated biphenyl (PCB) congener fingerprint data from the 2003–2004 National Health and Nutrition Examination Survey (NHANES). This approach enables interpretation beyond concentration-based biomonitoring by examining PCB congener fingerprints across demographic groups such as age, sex, and ethnicity in the U.S. population. Among the methods evaluated, UMAP provided the clearest low-dimensional representation of PCB congener fingerprints, revealing age-related differences in fingerprint structure, including increased contributions from heavier congeners among older individuals. UMAP further identified a distinct PCB congener fingerprint in a subset of the youngest cohort that did not follow the age-associated fingerprint patterns observed in this study or commonly reported in the literature, highlighting greater variability among younger individuals than is typically captured in studies measuring fewer PCB congeners than NHANES. This PCB congener fingerprint was characterized by the co-occurrence of elevated PCB-28 and PCB-209 relative to other age groups, distinguishing it from the dominant age-related patterns. While age-related differences in PCB congener fingerprints were evident, no discernible fingerprint differences were observed across sex or ethnicity groups. The limitation of the NHANES dataset to 35 predominantly Aroclor-related congeners underscores the importance of expanded congener coverage to fully characterize atypical or nontraditional PCB congener fingerprints. Overall, UMAP-based visualization provides a practical framework for comparing individual PCB congener fingerprints against typical demographic patterns and for identifying deviations that warrant further investigation within population biomonitoring datasets.