# Detection of Fluconazole Resistance in Candida parapsilosis Clinical Isolates with MALDI-TOF Analysis: A Proof-of-Concept Preliminary Study

## TL;DR

This study explores using mass spectrometry and machine learning to detect fluconazole resistance in Candida parapsilosis clinical isolates.

## Contribution

A proof-of-concept method using MALDI-TOF and machine learning for rapid detection of fluconazole resistance in Candida parapsilosis.

## Key findings

- Support Vector Machine with principal component analysis achieved 100% sensitivity in detecting fluconazole resistance.
- The method was tested on 39 isolates, including 16 susceptible and 23 resistant strains.
- External validation with more isolates from multiple centers is needed to confirm the results.

## Abstract

In the context of evolving antifungal resistance and increasing reports of clinical outbreaks of non-albicans Candida spp. invasive infections, the rapid detection of resistant patterns is of the utmost importance. Currently, an azole-resistant Candida parapsilosis clinical outbreak is ongoing at Pisa University Hospital. Resistant isolates bear both Y132F and S862C amino acid substitutions. Based on the data and isolates retrieved during the clinical outbreak, mass spectrometry was used to investigate the differences between fluconazole-resistant and -susceptible clinical strains directly from yeast colonies isolated from agar culture media. A total of 39 isolates, 16 susceptible and 23 resistant, were included. Spectra were processed following a standardized pipeline. Several supervised machine learning classifiers such as Random Forest, Light Gradient Boosting Machine, and Support Vector Machine, with and without principal component analysis were implemented to discriminate resistant from susceptible isolates. Support Vector Machine with principal component analysis showed the highest sensitivity in detecting fluconazole resistance (100%). Despite these promising results, external prospective validation of the algorithm with a higher number of clinical isolates retrieved from multiple clinical centers is required.