Abstract

Salmonella enterica is a leading cause of foodborne illnesses globally, with significant mortality rates, especially among vulnerable populations. Traditional serotyping methods for Salmonella are accurate but expensive, resource-intensive, and time-consuming, necessitating faster and more reliable alternatives. This study evaluates the IR Biotyper, a Fourier-transform infrared spectroscopy system, in differentiating Salmonella serovars. We assessed 458 isolates of nine Salmonella serovars (Infantis, Enteritidis, Typhimurium, I,4,[5],12:i:-, Montevideo, Agona, Thompson, Panama, and Abony) from diverse sources. The IR Biotyper was used to acquire spectra from these isolates. Machine learning algorithms, including support vector machines, were trained to classify the isolates. The accuracy of classifiers was validated using a validation set to determine sensitivity, specificity, positive predictive value, and negative predictive value. Initial classifiers showed high accuracy for Abony, Agona, Enteritidis, and Infantis serovars, with sensitivities close to 100%. However, classifiers for S. Typhimurium, S. Panama, and S. Montevideo exhibited lower performance. Implementing a hierarchical classification system enhanced the accuracy of serogroup O:4 serovars, demonstrating that this approach offers a robust framework for Salmonella serovar identification. The hierarchical system enables progressive refinement of classification, minimizing misclassifications by focusing on serogroup-specific features, making it adaptable to complex data sets and diverse serovars. The IR Biotyper demonstrates high potential for rapid and accurate Salmonella serovar identification. This study supports its implementation as a cost-effective, high-throughput tool for pathogen typing, enhancing real-time epidemiological surveillance, and guiding treatment strategies for salmonellosis. This method establishes a robust and scalable framework for advancing Salmonella serotyping practices across clinical, industrial, and public health domains by leveraging hierarchical classification. © © 2025 Fredes-García et al.