Machine learning assisted identification of antibiotic-resistant strains using a paper-based ratiometric sensor array.

, a versatile human pathogen, significantly impacts global health causing a broad spectrum of medical conditions that range from minor skin infections to life-threatening diseases. The clinical importance of is underscored by its resistance to multiple antibiotics and formation of biofilms, providing protection against antimicrobials and immune responses. To date, the identification of antimicrobial-resistant (AMR) strains, such as methicillin-resistant (MRSA) and vancomycin-intermediate (VISA), requires time-consuming and expensive methodologies, including culture-based, molecular, and phenotypic techniques. Previously, we developed a paper-based ratiometric sensor array composed of fluorescent sensor dyes (3-hydroxyflavone derivatives) pre-adsorbed on paper microzone plates. Combined with machine learning algorithms such as neural networks, this sensor effectively discriminated 16 bacterial species and determined their Gram status. In this study, we evaluate its ability to distinguish antibiotic-resistant strains and their biofilms. Our results demonstrate that the sensor array, in conjunction with LDA and neural networks, successfully differentiated three common laboratory MRSA strains from three methicillin-susceptible (MSSA) strains with 82.5% accuracy. Furthermore, using support vector machines, this sensor was able to distinguish and categorically classify MRSA, MSSA, and VISA clinical isolates with 97.5% accuracy. Remarkably, beyond distinguishing planktonic cultures, this sensor array demonstrated a formidable capability to discriminate AMR biofilms, achieving over 80% accuracy. Combined, the results of this study highlight the paper-based sensor array's significant potential as a robust diagnostic tool to accurately, rapidly, and easily identify drug-resistant strains in clinically relevant settings.