Artificial Intelligence to Combat Antimicrobial Resistance: Comparative Benchmark of Predictive Methods
Negar Ahvar, Mohsen Mohammadi, Pourya Zafari, Mehran Shokri · InfoScience Trends · 2025
Antimicrobial resistance (AMR) is a critical global health threat, necessitating advanced approaches for prediction, surveillance, and intervention. Artificial intelligence (AI) and machine learning (ML) offer promising tools for analyzing large-scale genomic, clinical, and chemical data to combat AMR. However, there is a lack of systematic, quantitative comparisons of AI model performance across different AMR applications. This study aimed to conduct a review and benchmarking analysis of AI models applied to AMR, comparing the performance of different model families across key domains, including genomic and phenotypic resistance prediction, clinical decision support, drug discovery, rapid diagnostics, and surveillance. We performed a review of empirical studies published from 2010 onward that employed AI/ML models for AMR-related tasks. Eligible studies reported quantitative performance metrics such as AUROC, accuracy, or MIC regression errors. We extracted data on model types, data modalities, pathogens, antibiotics, and performance outcomes. A benchmarking framework was applied to compare model families—including tree-based ensembles, deep learning architectures, linear models, and kernel-based methods—stratified by application domain and data type. Tree-based ensemble methods, such as random forests and gradient-boosted trees, consistently demonstrated strong performance in genomic and clinical prediction tasks, often outperforming linear models. Deep learning models showed domain-specific advantages, particularly in tasks involving high-dimensional data such as whole-genome sequences, spectroscopic inputs, and molecular graphs, but did not exhibit uniform superiority across all settings. In areas such as drug discovery, rapid antimicrobial susceptibility testing, and surveillance, model performance was highly dependent on data representation and quality, with limited evidence for clear superiority of any single model family. AI models, particularly tree-based ensembles, are effective tools for AMR prediction and management, though performance is context-dependent. Future work should focus on standardized benchmarking, improved data sharing, and real-world validation to translate AI advancements into measurable impacts on antimicrobial stewardship and resistance outcomes.