A Multitarget QSAR Approach for Virtual Screening of Natural Product and Drug Databases to Identify Novel Antifungals for Deep Mycose

Deep mycoses are a type of fungal infection considered a neglected tropical disease by the World Health Organization (WHO), affecting at least hundreds of thousands of people around the world. While there are treatments for fungal diseases, the rise of antifungal resistance urges the research of novel antifungal agents. This study aims to discover new molecules from natural products and repurpose clinically evaluated drugs that present potential antifungal activity against deep mycoses pathogens, through the construction and deployment of machine learning models based on multitarget quantitative structure-activity relationships (QSAR). Using experimental IC50 data from the ChEMBL database, 40 machine learning algorithms were tested, and the 5 best performing algorithms (Bagging, Gradient Boosting, LightGBM, Random Forest, XGBoost) were used to predict the antifungal activity of nearly 460,000 compounds from natural products as well as repurposing databases . This analysis resulted in the selection of 57 compounds considered active by consensus of the 5 models, including beta-lactams, griseofulvin derivatives and peptides, as well as standard antifungal drugs, such as amphotericin B and Nystatin, demonstrating the capability of the models to identify antifungal activity. Furthermore, the calculaton of SHAP values provided mechanistic insight, confirming that the models' predictions were driven by molecular substructures characteristic of established antifungal agents. This study highlights potential antifungal agent candidates, both novel and repurposed, warranting further in silico, in vitro, and in vivo studies.