COMPUTATIONAL MODELING FOR PREDICTING ANTIBIOTIC RESIDUES RISK IN AQUATIC ECOSYSTEMS

Madhushree M B, Amrutha V N, Sumedha P, Nayana G Singh, Kushan T S

Due to the horizontal transmission of antibiotic resistance genes among microbial communities in surface waters, sediments, and groundwater, the proliferation of antibiotic residues in aquatic ecosystems has become a major environmental concern. Empirical monitoring programs are limited by high costs, spatiotemporal variability, and the combinatorial complexity of multi- contaminant matrices, despite the fact that they offer essential occurrence data. By facilitating systematic risk assessment across spatial dimensions and exposure scenarios not possible through traditional field campaigns, computational predictive modeling provides an affordable supplement. In order to critically assess the state of quantitative structure–activity relationship models, machine learning algorithms, and mechanistic fate-and-transport frameworks applied to antibiotic risk prediction in freshwater and coastal environments. The findings show that while mechanistic models are excellent at capturing nonlinear hydrological drivers but necessitate extensive site-specific parameterization, ensemble machine learning models— especially gradient-boosted trees and graph neural networks—consistently outperform classical quantitative structure–activity relationship models approaches (a mean R² improvement of 12–18%). Although they are still unexplored, hybrid modeling approaches that combine machine learning with process-based hydrological equations show great promise. The main findings emphasize the necessity for standardized environmental input datasets, the ongoing limitations of data scarcity, and the regulatory significance of uncertainty quantification in model outputs. Protocols for model benchmarking and approaches for policy integration are suggested.

Keywords –Antibiotic residues; antibiotic resistance genes; aquatic risk prediction; environmental computational modeling; machine learning; mechanistic fate models; Quantitative Structure Activity Relationship.