Quantitative Structure-Activity Relationship Models for Cryptococcosis: A Virtual Screening Approach

Clinical treatment of cryptococcal meningitis (CM) remains a significant challenge because of the lack of effective and safe drug therapies. Developing novel CM therapeutic agents is of great importance. In this study, we report the development of binary QSAR models useful for virtual screening1 of novel antifungal compounds. Initially, 4,607 chemical structures with MIC data for Cryptococcus neoformans were curated and standardized according to the protocol proposed by Fourches and coworkers.2 Then, binary QSAR models combining Morgan and FeatMorgan fingerprints with three machine learning methods (Support Vector Machine, Random Forest, and XGBoost) were developed following the best practices of QSAR modeling.3 As a result, we have obtained statistically predictive QSAR models, with CCR, Sensitivity, and Specificity values ranging between 0.82–0.86. Finally, we applied the best models for virtual screening of eMolecules library, which allowed us to select 20 virtual hits for prospective in vitro assays.