Akademik Veri Yönetim Sistemi
MicrobiologyOpen, cilt.14, sa.5, 2025 (SCI-Expanded)
Candidozyma (Candida) auris is a multidrug-resistant yeast capable of causing persistent outbreaks in healthcare settings. Its reduced susceptibility to both antifungals and biocides poses a direct threat to infection control and hospital hygiene, as biocides remain the cornerstone of surface disinfection and outbreak containment. This study aimed to construct a decision tree to predict biocide susceptibility and identify key predictive parameters using machine learning. Virulence factors were evaluated in 55 C. auris strains isolated from a state hospital in Türkiye that were identified by MALDI-TOF and verified by DNA sequencing. CLSI guidelines were followed in determining the antifungal MICs. The z-score method was used to standardize and numerically code phenotypic data. The Random Forest Regressor was used to analyze feature importance. Resistance thresholds were defined as triclosan ≥ 0.5 µg/mL, benzalkonium ≥ 150 µg/mL, chlorhexidine ≥ 1.0 µg/mL, and chlorine ≥ 0.03 µg/mL. As all strains were benzalkonium-resistant, these data were excluded. Anidulafungin MIC was the strongest predictor of biocide sensitivity, followed by amphotericin B, flucytosine, and isavuconazole, while other virulence factors showed little or no value. This proof-of-concept study demonstrates, for the first time, that a decision tree model trained on antifungal MIC profiles can predict the susceptibility of C. auris strains to triclosan, chlorine, and chlorhexidine. Although biocide susceptibility testing was performed to establish reference thresholds, the final predictive framework relied solely on anidulafungin MIC values, suggesting that such models may reduce the need for routine biocide testing in future surveillance studies.