Computational paradigms for antimicrobial resistance prediction: integrating multi-omics, structural modeling, and...

Area of Science:

• Genomic epidemiology
• Computational biology
• Machine learning in infectious diseases

Background:

• Antimicrobial resistance (AMR) is a growing global health crisis, threatening medical treatments and surveillance.
• Current multi-omics data analysis for AMR lacks clinical integration due to computational fragmentation.

Purpose of the Study:

• To systematically review and synthesize machine learning and AI approaches for antimicrobial resistance (AMR) modeling.
• To identify key methodological trends and challenges in translating multi-omics data into clinically actionable AMR intelligence.

Main Methods:

• Systematic literature review following PRISMA 2020 guidelines (156 records screened, 93 studies synthesized).
• Categorization of AMR modeling into classical ML, deep genomic architectures, and transformer/LLM systems.
• Analysis of integrative directions: multimodal fusion, knowledge graphs, forecasting, and agentic AI.

Main Results:

• AMR modeling encompasses classical ML, deep learning, and advanced AI (transformers, LLMs) integrating diverse data.
• Converging research trends include multimodal data fusion, causal reasoning, temporal forecasting, and autonomous AI workflows.
• Significant heterogeneity exists in dataset scale, validation methods, and robustness, especially for rare resistance phenotypes.

Conclusions:

• Future AMR intelligence requires uncertainty-aware models, standardized validation, and FAIR data principles.
• Transitioning from static classification to adaptive, interpretable, and clinically actionable decision systems is crucial for One Health surveillance.