Shapley Additive Explanations of Indicator PCB-138 Distribution in Breast Milk

Breastfeeding provides numerous health benefits for newborns by meeting the infantsʼ nutritional needs and supporting associated immunological protection. Maternal milk is high in fat, and therefore, represents a very suitable media for the bioaccumulation of lipophilic pollutants such as organochlorine pesticides (OCPs) and polychlorinated biphenyls (PCBs). This makes breast milk the primary source of the infant’s postnatal exposure to persistent toxic xenobiotics. In this study, we applied a novel SHapley Additive exPlanations (SHAP) method to investigate the key parameters that govern distribution of PCB-138, an indicator of non-dioxin congeners, in the mother’s milk. According to the accuracy metrics, the eXtreme Gradient Boosting regression was successfully employed, with relative error below 20% and high correlation coefficient (r=0.97), for finding the relationships between PCB-138 and other non-dioxin congeners, mother’s age, and number of births. According to the results, PCB-156, PCB-180, HCB, HCH and PCB-118 have the major, while PCB-28, PCB-52 and PCB-189 have the minor impact on PCB-138 distribution in breast milk. Similar behavior of contaminants, which belong to the both indicator congener group (−28, −52, −180) and toxicologically relevant PCBs (−118, −189), was recognized. The SHAP conclusions were only partially consistent with the results of correlation analysis suggesting that POPs exhibit non-linear dynamics and interrelations. Therefore, the current knowledge on contamination of complex biomatrices would benefit from further detailed analyses of pollutant intermittent relationships.