Multimodal Raman-Gene Framework for Interpretable Phenotypic Antibiotic Resistance and Health Risk Stratification

The increasing burden of antibiotic-resistant bacteria presents a major challenge for clinical management, as conventional antimicrobial susceptibility testing (AST) depends on culture-based workflows that are slow, labor-intensive, and poorly suited for rapid clinical decision-making. Moreover, standard AST primarily reports susceptibility phenotypes, while providing little insight into the potential health risks associated with resistance dissemination and pathogenicity. Here, we develop a culture-free multimodal Raman-gene-deep learning strategy for direct phenotypic antibiotic resistance profiling and resistance risk assessment in urine samples. By integrating surface-enhanced Raman spectroscopy (SERS)-derived phenotypic fingerprints with targeted genetic information on antibiotic resistance genes and virulence factors, this approach enables resistance characterization directly from clinical samples without bacterial isolation and culture. Importantly, in addition to rapid phenotypic antibiotic resistance prediction, the proposed approach enables health risk assessment of antibiotic-resistant bacteria by integrating indicators related to clinical impact, transmission risk, and pathogenicity. Applied to clinical urine samples, the method delivers accurate phenotypic resistance results within approximately 2.5 h and simultaneously provides risk-level information that is not available from routine culture-based AST. Overall, this study presents a proof-of-concept framework for rapid phenotypic resistance prediction and risk assessment, offering enhanced informational depth, reduced operational complexity, and potential utility for antimicrobial decision-making and infection control.