Genome analysis and machine learning-based feature selection strategy reveal potential drug-resistance determinants...

Invasive candidiasis caused by Nakaseomyces glabratus is of great concern due to high morbidity and mortality, especially antifungal resistance. To identify genomic signatures, which significantly link to drug-resistance, is of great significance in combating this lethal disease. In this study, we performed whole genome analysis on 109 clinical strains of N. glabratus which had been isolated from multi-centres in China. By using genome-wide association studies (GWAS), genomic signatures, including several PDR1 mutations and genes encoding GLEYA-containing proteins, were identified to be significantly linked to drug-resistance. With the strategy of feature-selection combining machine-learning (ML), more relevant genomic signatures and potential resistance determinants were identified, including Y682C and I380L mutations in PDR1 which were further confirmed to confer triazole-resistance by gene editing technology. We believe that the ML-based feature selection (MLFS) strategy, which is based on a comprehensive understanding of genomic characteristics as described in this study, shows excellent capacity to predict resistance and potential resistance determinants in N. glabratus.