Integrated assessment of developmental toxicity of antibiotic pollutants: Machine learning prediction, zebrafish validation, and network toxicology.

Developmental toxicity induced by environmental pollutants, particularly antibiotics, is often insidious and underestimated due to bioaccumulation and subsequent oral intake. This study developed a machine learning-based predictive strategy by constructing and comparing four models using Morgan fingerprints on a curated dataset of developmental toxicants. The optimal random forest model achieved an area under the receiver operating characteristic curve (AUC) of 0.87 (95% confidence interval: 0.83-0.91), an accuracy of 0.80, and a Matthews correlation coefficient of 0.59. Applying the ensemble model to 2,341 antibiotics identified miconazole as the highest-probability candidate (average probability 0.97). In zebrafish embryo toxicity assays, exposure to miconazole at low concentrations (0.3 and 3.0 µM) did not result in statistically significant mortality up to 72 h post-fertilization, whereas the high concentration (30 µM) caused significantly elevated mortality at 48 and 72 h (p < 0.05 compared to control). Network toxicology and molecular docking revealed that miconazole may interact with key targets AKT1 and BRAF, potentially perturbing the chemical carcinogenesis-reactive oxygen species signaling pathway. These integrated findings indicate that miconazole exhibits potential developmental toxicity, warranting further mechanistic and long-term exposure studies before drawing definitive regulatory conclusions.