Detection of Fluconazole Resistance in Candida parapsilosis Clinical Isolates with MALDI-TOF Analysis: A...

Area of Science:

• Clinical Microbiology
• Mycology
• Computational Biology

Background:

• Rising antifungal resistance and invasive non-albicans Candida infections necessitate rapid detection methods.
• An ongoing azole-resistant Candida parapsilosis outbreak at Pisa University Hospital highlights this urgent need.
• Resistant isolates in the outbreak harbor specific amino acid substitutions (Y132F and S862C).

Purpose of the Study:

• To differentiate fluconazole-resistant from susceptible Candida parapsilosis clinical strains using mass spectrometry.
• To evaluate the efficacy of machine learning classifiers for rapid antifungal resistance detection.
• To leverage data from a current clinical outbreak for method development.

Main Methods:

• Mass spectrometry was employed to analyze 39 Candida parapsilosis isolates (16 susceptible, 23 resistant) directly from colonies.
• Standardized spectral processing pipeline was applied.
• Supervised machine learning classifiers (Random Forest, Light Gradient Boosting Machine, Support Vector Machine) with and without Principal Component Analysis were implemented.

Main Results:

• Support Vector Machine with Principal Component Analysis achieved 100% sensitivity in identifying fluconazole resistance.
• Machine learning models demonstrated the potential for rapid discrimination between resistant and susceptible strains.
• Distinct spectral profiles were observed between resistant and susceptible isolates.

Conclusions:

• Mass spectrometry combined with machine learning, particularly Support Vector Machine with PCA, offers a highly sensitive method for detecting fluconazole resistance in Candida parapsilosis.
• This approach shows promise for rapid identification of antifungal resistance patterns during clinical outbreaks.
• Further external prospective validation with larger, multi-center datasets is required to confirm the algorithm's generalizability.