Artificial intelligence in microbiology: implications for metagenomics, diagnostics, and AMR surveillance

Artificial intelligence (AI) is now a key player in modern microbiology, as it enables high-resolution analyses of genomic, metagenomic, and clinical data for the monitoring of infectious disease and antimicrobial resistance (AMR). Considerable advancements in deep learning, transformer-based sequence models, graph neural networks, and multimodal architectures have greatly improved microbial classification accuracy, antibiotic resistance gene (ARG) detection, and resistance prediction. Taking metagenomic sequencing into consideration, these advancements have contributed to the development of sensitive, scalable, and non-invasive methods to profile microbiomes, determine novel resistance, and monitor AMR trends at the population level. This review summarizes recent advances in AI-aided microbiology, with a particular emphasis on AMR surveillance. Specific topics include deep learning frameworks for ARG annotation, emerging approaches to identifying new resistance genes, and multimodal applications (genomic and clinical metadata) aimed at improving phenotype prediction. The role of metagenome-assembled genomes (MAGs) to enhance AMR surveillance efforts is noted, along with their noted limitations relative to isolate genomes. The discussion includes the examination of explainable AI (XAI) techniques including SHAP, attention mechanism approaches, and gradient-based attribution approaches, with the aim of increasing transparency and clinical explainability. We also cover potential applications including AI-enabled non-invasive fecal microbiome diagnostics, laboratory automation, and environmental surveillance. While there has been significant progress, unresolved issues exist relating to dataset variations, liability of models to datasets, interpretability, and regulatory approval. Overcoming these barriers, however, will require standardized frameworks for these workflows, privacy-preserving federated learning methods, and interpretable AI frameworks for clinical and public health tools. AI could fundamentally change AMR surveillance by allowing for earlier resistance detection, advanced risk assessment recommendation, and improved monitoring strategies globally.