Highly Efficient Broad-Spectrum Antibacterial Carbon Dots through Hierarchical Machine Learning Framework

The global threat of antibiotic resistance necessitates intelligent design strategies for next-generation antibacterial nanomaterials. Herein, high-efficacy broad-spectrum antibacterial carbon dots (CDs) are demonstrated by the developed hierarchical machine learning (ML) framework. A classification ML model is first employed to screen CDs by antibacterial type, followed by a regression model to predict and optimize bactericidal efficacy. The resulting CDs exhibit 99.99% bactericidal efficacy against both Gram-positive and Gram-negative pathogens under 660 nm irradiation. Their positively charged surfaces (+25 mV) facilitate targeted interactions with bacterial membranes, while in situ reactive oxygen species (ROS) generation enables efficient bacterial inactivation. SHapley Additive exPlanations (SHAP) analysis reveals that positively charged and hydrophilic-dominated CD surfaces are key determinants of efficient and broad-spectrum antibacterial performance. Ultimately, ML-designed CDs demonstrate excellent therapeutic efficacy in a murine model of bacterial wound infection. This work highlights the potential of hierarchical ML-assisted strategies for developing highly efficient broad-spectrum antibacterial nanomaterials.