Pilot study of Raman-deep learning integration for rapid detection of antibiotic-resistant Escherichia coli: A proof-of-concept analysis

In recent years, the accelerating emergence of antibiotic-resistant bacterial pathogens has posed profound challenges to the therapeutic management, prophylactic strategies, and epidemiological control of infectious diseases caused by these microorganisms. Consequently, identifying resistant bacteria is essential for therapeutic decisions and epidemiological studies. However, conventional approaches for the detection of antibiotic resistance are frequently constrained by lengthy protocols, substantial costs, and operational intricacies, thereby impeding the rapid and precise identification of resistance phenotypes. Raman optical tweezers have proven useful for classifying different bacterial species and isolates. This study establishes a fast, reliable, and cost-effective method to differentiate between Escherichia coli strains with antibiotic resistance (extended spectrum β-lactam resistant, ESBL) and susceptible strains using Raman optical tweezers and deep learning techniques. High-quality single-cell Raman spectra were collected from antibiotic-resistant and susceptible strains without exposure to antibiotics, revealing a higher nucleic acid/protein ratio in resistant strains. We propose a new network, RamanU-Net, for the accurate classification of these Raman spectra. The model achieved 99.5% average accuracy in identifying antibiotic-resistant and sensitive strains of Escherichia coli. Our results demonstrate that combining Raman optical tweezers with deep learning models can enable rapid identification of bacterial antibiotic resistance while significantly reducing the associated time, cost, and workload.