Unveiling latent interaction patterns of heavy metals in soils using Bayesian network structure learning: Study from...

Understanding the interaction patterns among heavy metals in contaminated soils in industrial regions remains challenging, as traditional correlation-based methods fail to reveal the directional dependencies underlying co-mobilization processes. This study aimed to unveil latent causal pathways among nine heavy metals (As, Co, Cr, Cu, Fe, Mn, Ni, Pb, Sb, and Zn) in surface soils from an industrialized region in Central China using Bayesian network (BN) structure learning. The concentrations of Cu (1.10 to 1247 mg kg^-1), Sb (0.040 to 5.67 mg kg^-1), Pb (4.31 to 204 mg kg^-1), Zn (0.329 to 352 mg kg^-1), and As (0.166 to 61.8 mg kg^-1) showed marked spatial variability across the study area. Using BDeu scoring with bootstrap validation, eight high-confidence edges were detected, forming two dominant pathways: Fe → Cr → Co → Pb/Cu and Mn → Co → Pb/Cu, respectively. These pathways indicate that redox-sensitive elements organize multi-metal interaction structures by regulating upstream transmission processes, which govern downstream accumulation patterns. The results differentiate metals that primarily drive propagation from those that act as convergence endpoints, revealing an interaction hierarchy that cannot be captured by enrichment or correlation analysis alone. Two functional clusters were further identified: a process-driven transmission group (Cr, Fe, Co, Sb) and a heterogeneous convergence group (Ni, Cu, Mn, Pb, As, Zn). Despite significant enrichment, As and Zn remained structurally isolated, indicating point-source contamination. Multi-pathway convergence analysis revealed that 55% of industrial samples and 27% of residential samples activated over 5 transmission chains simultaneously, forming hotspots with cumulative mobilization risk. Findings demonstrate BN-inferred directional dependencies provide mechanistic intelligence for prioritizing upstream control of high-centrality nodes, offering a robust framework for risk assessment and targeted remediation in complex contaminated landscapes.