Causal machine learning with interpretability deciphers the impact of micropollutants and socioeconomic factors on ARGs in Chinese urban drinking water.

The widespread co-occurrence of antibiotic resistance genes (ARGs) with diverse micropollutants in drinking water distribution systems poses a critical public health threat by potentially facilitating ARG dissemination, yet the underlying causal drivers remain poorly understood. In this study, we employed H2O automated machine learning to profile ARGs and nearly 100 micropollutants, alongside socioeconomic indicators, in drinking water samples collected from 67 Chinese cities. Moving beyond traditional correlation and interpretability analyses, we adopted double machine learning (DML), a causal inference framework, to quantified causal effects (CATE) and elucidate pollutant-ARG relationships. Results revealed synergistic effects among major ARG types (e.g., sul1, tetB and blaTEM), with integron genes (intI1, intI2) serving as key genetic vectors. Antibiotics (e.g., Sulfaphenazole) and PAHs (e.g., Acenaphthene) significantly drove the proliferation of ARGs, while PCBs (e.g., Perfluoro-n-dodecanoic acid) and advanced urban development generally suppressed their prevalence, with GDP exhibiting a nonlinear U-shaped association in detail. The integration of interpretable machine learning with DML effectively deciphered these intricate relationships. Our findings provide new causal insights into ARGs drivers in drinking water and supports evidence-based risk assessment and targeted strategies to mitigate antimicrobial resistance dissemination via water systems.