Development and validation of a hierarchical machine learning method using MALDI-TOF mass spectrometry for rapid SCCmec typing and PVL detection in MRSA: a multi-centre study

Abstract

Objectives: Methicillin-resistant Staphylococcus aureus (MRSA) is a major public health concern because of its genotypic diversity and association with severe infections, particularly those caused by strains carrying Panton-Valentine leukocidin (PVL). This study aimed to develop an artificial intelligence-clinical decision support system (AI-CDSS) to streamline MRSA genotyping and PVL detection, providing a more efficient alternative to complex PCR-based workflows.

Methods: We retrospectively analysed 345,748 bacterial specimens collected from five healthcare institutions between 2010 and 2024. Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry data were analysed using a hierarchical classification framework enhanced by machine learning models to identify the MRSA status, staphylococcal cassette chromosome mec subtypes, and PVL presence. Area under the curve (AUC), sensitivity, and specificity were used for model evaluation.

Results: AI-CDSS was highly accurate for MRSA genotyping (AUCs > 0.9) and PVL detection (AUC = 0.85). Automating hierarchical classifications effectively replaced labour-intensive PCR processes, reducing diagnostic complexity and resource use.

Conclusions: AI-CDSS is a scalable and efficient method for MRSA genotyping and PVL detection. By streamlining diagnostics and supporting timely clinical interventions, this system can improve infection management and patient care, which will reduce mortality associated with MRSA infections.

Keywords: AI-CDSS; MRSA; PVL; artificial intelligence; artificial intelligence-clinical decision support systems.

Figure 1.

Overview of the MRSA genotyping and PVL detection workflow, integrating traditional methods with AI-CDSS-enhanced processes. The traditional workflow begins with clinical specimen collection on Day 0, followed by pathogen identification using MALDI-TOF MS on Day 3, antibiotic sensitivity testing, and MRSA/MSSA classification on Day 4. Next-generation sequencing is also performed for further strain characterization. The AI-CDSS system leverages MALDI-TOF MS data, enhancing the diagnostic process through hierarchical classification. It first distinguishes between MRSA and MSSA, then categorizes MRSA into hospital-acquired (HA-MRSA) and community-acquired (CA-MRSA) strains, followed by SCCmec typing and PVL status detection. This AI-enhanced system streamlines the workflow, improving diagnostic speed and enabling more efficient, timely clinical decision-making. MRSA, methicillin-resistance Staphylococcus aureus; MSSA, methicillin-sensitive Staphylococcus aureus; PVL, Panton–Valentine leucocidin; AI-CDSS; artificial intelligence-clinical decision support system; MALDI-TOF MS, matrix-assisted laser desorption/ionization time-of-flight mass spectrometry; SCCmec, staphylococcal cassette chromosome mec

Figure 2.

Correlation heatmap of the top 10 m/z segments and PVL + status. The heatmap visualizes the correlation coefficients between selected m/z segments and the presence of PVL (PVL+) in the dataset. The colour intensity represents the strength and direction of the correlation, with red indicating a strong positive correlation and blue indicating a weak or negative correlation. Notably, the m/z segments 4625–4630 and 4660–4665 show a strong positive correlation with PVL+, while other m/z ranges, such as 4285–4290 and 4050–4055, exhibit weaker associations. This analysis highlights the m/z segments that are most predictive of PVL presence.

Figure 3.

MRSA AI-CDSS web interface. The web interface displays the stepwise output of a hierarchical machine learning model for Staphylococcus aureus typing based on MALDI-TOF MS data. The AI Hierarchical Decision Path presents each classification step, along with the predicted outcome at each stage. The Confidence for Each Step section summarizes the model’s probability score for each decision. The Diagnostic Summary card provides a concise overview of the final molecular characteristics and corresponding confidence scores, facilitating clinician interpretation and informed decision-making. MRSA, methicillin-resistance Staphylococcus aureus; AI-CDSS; artificial intelligence-clinical decision support system; MALDI-TOF MS, matrix-assisted laser desorption/ionization time-of-flight mass spectrometry.