Patterns of PCB-138 Bioaccumulation in Small Pelagic Fish from the Eastern Mediterranean Sea Using Explainable Machine Learning Prediction

Abstract

Fish consumption, especially consumption of oily marine species, is increasing globally due to its recommendation by dieticians. This is due to high polyunsaturated $\omega$-3 and $\omega$-6 (PUFAs) fatty acid content in the tissue of the fish. The health benefits of PUFA ingestion coincide with the risk of intaking hazardous lipophilic persistent pollutants including organochlorine pesticides (OCPs) and related polychlorinated biphenyls (PCBs). We examined the impact of 17 fatty acids (FAs) and 36 toxic organic and inorganic contaminants on the behavior patterns of the indicator congener PCB-138 in marine fish using eXtreme Gradient Boosting (XGBoost), SHapley Additive exPlanations (SHAP), and SHAP value fuzzy clustering. XGBoost indicated non-linear relationships between PCB-138 and other investigated variables that were explained by SHAP values. The ten obtained fuzzy clusters of SHAP values revealed that a higher intake of saturated myristic-C14:0 and margaric-C17:0 acids followed by the intake of nutritionally beneficial eicosadienoic acid (C20:2n-6) mostly do not increase the bioaccumulation of PCB-138. Important effects on PCB-138 behavior patterns were also recorded for the chemically allied indicator congeners ($-153$, $-180$, $-118$ and$-101$) and organochlorines’ metabolite p,p’-DDE. Associations between the target congener and the toxicologically relevant PCBs ($-123$ and $-170$) were less prominent.