As interest continues in bacteriophages as precision antimicrobials, understanding the molecular determinants of host range has become more important. Here, we use the term polyvalent to describe phages whose broad host recognition is associated with more than one receptor-binding protein (RBP) and/or more than one experimentally supported receptor-recognition pathway, which is distinct from broad host range as a phenotypic description alone. We propose a stochastic-cooperative hypothesis in which some RBPs can contribute primarily to reversible surface encounter and retention, whereas other RBPs mediate the stable receptor binding required to progress toward genome delivery. The novelty of this hypothesis is not the established reversible-to-irreversible sequence of adsorption, nor the existence of phages with multiple RBPs, but the specific prediction that functionally non-equivalent RBPs can alter the probability distribution of transitions between adsorption states on heterogeneous bacterial surfaces. Available structural, genetic, and single-particle evidence is consistent with parts of this interpretation but does not currently demonstrate a universal auxiliary-first pathway. Therefore, we distinguish evidence-supported elements from hypothetical components, consider alternative explanations for broad host range, and define experimentally testable predictions at molecular, single-particle, cellular, and population scales. This Perspective is intended as an experimentally evaluable conceptual model for studying how receptor density, accessibility, RBP composition, and post-adsorption barriers jointly shape observed host range, rather than as an established mechanism for all polyvalent phages.

Viral metagenomics, leveraging high-throughput sequencing technologies, provides comprehensive, hypothesis-free characterization of viral communities in clinical specimens, establishing itself as a pivotal tool for clinical diagnosis, pathogen discovery, and epidemiological surveillance of viral infectious diseases. The integration of artificial intelligence (AI) has demonstrated transformative potential in viral metagenomic data analysis, significantly enhancing sequence classification, feature extraction, pattern recognition, and result interpretation, thereby improving analytical efficiency, accuracy, and automation. Machine-learning and deep-learning approaches have shown potential for detecting low-abundance viral signals and mitigating background noise in complex clinical samples, and facilitating pathogen identification in complex clinical samples, substantially augmenting the clinical utility of viral metagenomics. Nevertheless, routine clinical implementation faces persistent challenges, including high host nucleic acid background, low viral titers, contamination control, limited reference databases, absence of standardized analytical pipelines, and limited interpretability and generalizability of some predictive models. Furthermore, clinical validation, result reproducibility, ethical compliance, and data security represent critical translational barriers that impede widespread adoption. This review comprehensively summarizes recent advances and clinical applications of AI-empowered viral metagenomics, highlighting its transformative potential in pathogen detection, diagnosis of challenging infections, infectious disease surveillance, and precision medicine. We comprehensively address major technical bottlenecks and translational barriers that must be overcome to facilitate standardized development and clinical implementation of this emerging paradigm.