Machine Learning-Driven Metagenomics for Tracing Antimicrobial Resistance Gene Flux Across Agricultural, Wastewater, and Clinical Environments

Abstract

Antimicrobial resistance has become a significant One Health problem, as antimicrobial resistance genes can move through agricultural systems, wastewater networks, clinical settings, and natural ecosystems. Culture based and targeted molecular approaches are limited in their ability to provide information on the whole resistome, microbial hosts, mobile genetic elements and the directionality of resistance determinants movement. In this paper, we present AMR-FluxNet, a machine learning-based metagenomic framework to track the source, transmission direction, intensity, and future trajectory of antimicrobial resistance genes across agricultural runoff, wastewater, and clinical environments. The framework combines shotgun genome sequencing, taxonomic profiling, resistome characterization, mobilome analysis, ARG-host association, environmental metadata, and spatiotemporal graph attention learning. Sampling locations, microbial taxa, resistance genes, clinical isolates and mobile genetic elements are nodes in a dynamic heterogeneous graph with hydraulic, spatial, biological and temporal relationships modelled as weighted edges. A relation-aware graph attention network captures cross-environment interaction and a gated temporal encoder learns delayed and seasonal ARG movement. The framework also estimates future abundance, directional ARG flux, source contributions and a mobility-aware risk index using gene abundance, mobile-element association, pathogenic-host probability, clinical relevance, temporal growth and incoming flux. Illustrative evaluation results show that AMR-FluxNet achieves 94.6% source-attribution accuracy, a macro-F1 score of 93.6% and a (R\^{}2) value of 0.912 for ARG-flux prediction, outperforming conventional source-tracking, machine learning, recurrent and graph-based models. The proposed framework is scalable for early-warning surveillance, hotspot detection and specific measures across agriculture, wastewater treatment and health systems.