Metal bioaccumulation and lipid response in aquatic ecosystem: Evidence of machine learning from Chaohu Lake, China

Potentially toxic metals pose threats to ecosystems through bioaccumulation, yet the traceability of metal bioaccumulation pathways and their induction of metabolic disorders in fish tissues remain poorly understood. Here, we present an integrative study combining multi-media metal analysis, lipidomics, and machine learning to decipher metal-lipid interactions in Chaohu Lake, China-a critical freshwater system facing industrial and agricultural pollution. Analyzing 16 metals across water, sediments, and four fish tissues (muscle, liver, gonad, brain) and profiled 297 lipid species specifically in these fish tissues, we developed a SHAP-BF (Shapley Additive Explanations-Bioaccumulation Factor) framework to quantify metal origins and tissue-specific lipid responses. Our framework revealed dual water-sediment sources for brain/gonad metals (e.g., Mn, Ni) versus predominant water-derived accumulation in muscle/liver (Ni, Be, Mn, Co). Lipid profiling identified glyceride dominance in muscle/liver/gonad (64.7-69.1 %) and phospholipid enrichment in brain (75.4 %), with muscle lipids exhibiting the highest sensitivity to aqueous metals. Machine learning further linked Ni and Mn to disrupted phospholipid homeostasis (e.g., PC 39:6, PE 32:1 et al.) in muscle, highlighting risks to nutritional quality and food safety. This work pioneers an AI-driven approach for tracing metal transfer dynamics and lipidomic disruptions, offering a transformative framework for ecological risk assessment and pollution mitigation strategies.