Artificial Intelligence in Virus Detection: From Metagenomic Discovery to Clinical Decision Support and Population Surveillance

Artificial intelligence (AI) is increasingly central to modern virus detection, particularly as metagenomic next-generation sequencing (mNGS) expands diagnostic and surveillance capabilities beyond targeted assays. This paper reviews how AI supports virus detection across three interconnected levels. The metagenomic layer forms the technical core of the review, while the clinical and population sections examine how sequence-level AI outputs are translated into patient-level and public-health decision support. At the metagenomic level, AI addresses key bottlenecks caused by fragmented assemblies and weak homology signals. Foundation-model approaches such as ViraLM leverage pretrained genome representations to improve viral contig discrimination under short-contig conditions, whereas LucaProt integrates sequence- and structure-informed learning to detect highly divergent RNA viruses via RdRP (RNA-dependent RNA polymerase)-focused discovery. At the clinical level, targeted molecular tests remain effective when the suspected pathogen is known, while mNGS provides a complementary, hypothesis-free approach that can broaden etiological investigation in complex infections when laboratories apply appropriate quality control and clinicians interpret results in context. At the population level, early warning systems combine heterogeneous data sources (epidemiological, open-source, digital, environmental, and wastewater signals) with AI methods to enhance outbreak detection and forecasting. Across all levels, robust contamination control, transparent reporting, external validation, and standardized interpretation criteria determine real-world value. Overall, hybrid pipelines that integrate molecular evidence, clinical metadata, and population-scale surveillance within governed, reproducible workflows are likely to provide the greatest impact.