Utilizing machine learning and MALDI-TOF MS platform for rapid detection of ceftriaxone-resistant nontyphoidal Salmonella

(NTS) infections, but rising resistance threatens public health. Conventional antimicrobial susceptibility testing remains time-consuming, delaying treatment. Although MALDI-TOF MS enables rapid microbial identification, it lacks inherent resistance detection. This study integrates MALDI-TOF MS with machine learning to predict CRO resistance in NTS. Using 472 isolates (1,414 spectra) from three medical centers, we developed and validated a model based on the Light Gradient Boosting Machine (LGB) algorithm. With only seven features, LGB achieved AUCs of 0.94 (training), 0.88 (internal validation), 0.79 (external validation 1), and 0.80 (external validation 2). We further deployed the model as an interactive Streamlit web interface to demonstrate its potential usability and facilitate future laboratory evaluation. This approach provides a proof-of-concept framework for rapid preliminary prediction of ceftriaxone resistance in NTS using routinely generated MALDI-TOF MS spectra. However, further large-scale, prospective, multicenter validation is required before this model can be implemented as a routine clinical susceptibility testing tool. IMPORTANCE: in pediatric patients, particularly in cases requiring antimicrobial therapy, highlights the need for faster antimicrobial susceptibility testing. In this study, we developed a machine learning model integrated with MALDI-TOF MS to predict ceftriaxone resistance in NTS and deployed it as a web-based research-use interface. This proof-of-concept approach may support the future development of rapid adjunctive tools for antimicrobial resistance prediction.