Mapping Antibiotic Photocatalytic Transformation and Resistance Risks with a DFT-Informed Machine Learning Workflow

The photocatalytic degradation of antibiotics, while effective, risks generating transformation products (TPs) that may retain or amplify ecological threats, including antibiotic resistance gene (ARG) induction. This study addresses this challenge by developing a predictive, density functional theory (DFT)-informed machine learning (ML) framework that maps complex degradation networks and quantitatively assesses associated environmental risks. Using tetracycline as a model compound, we constructed an unprecedented reaction network of over 120 steps and 9,533 elementary reactions. To overcome the computational bottleneck of evaluating this network, an ML model was trained on DFT data to rapidly predict Gibbs free energy changes, enabling the first large-scale, network-wide thermodynamic feasibility assessment. This hybrid DFT-ML approach was successfully validated against the complex pathways of ciprofloxacin. Furthermore, a systematic pathway scoring system was developed to rank routes by integrating Diversity, Ecotoxicity, Biodegradability, and Feasibility of pathway (DEBF). Crucially, our analysis reveals that several TPs, particularly hydroxylated, amide ketone, and aminated species, pose a heightened risk of promoting antibiotic resistance. DFT-based docking studies confirm that these TPs form strong hydrogen bonds with key residues (Ile339A, Lys337A, Asp363A) in the tetM protein, potentially enhancing its binding affinity and ARG induction. This work provides a transformative strategy that bridges molecular-level reaction energetics with ecological risk assessment, guiding the development of photocatalytic water treatment processes that are not only efficient but also mitigate the unintended amplification of antimicrobial resistance.