Machine learning and orthology-based design of synergistic drug combinations against aspergillus fumigatus

Abstract

Background: Aspergillus fumigatus is a common fungal species that can cause invasive aspergillosis, particularly in immunocompromised individuals. In Uganda, around 9% of the population suffers from fungal diseases. Drug resistance and limited access to antifungals in low-income countries present major challenges. While drug development and vaccines offer potential solutions, they are often costly and time-consuming. To address this, the study utilized chemogenomic data from Cryptococcus neoformans and applied machine learning to identify synergistic drug combinations against A. fumigatus, using C. neoformans as a model and mapping orthologous genes to extrapolate drug interaction profiles to A. fumigatus. Methods: Large-scale datasets for Cryptococcus neoformans including chemogenomic data, antifungal drug structures, and Minimum Inhibitory Concentrations (MICs) from checkerboard assays were collected and used to train machine-learning models after performing a series of data preprocessing steps. Machine learning algorithms like random forest and gradient boosting classifiers and ensemble methods were used to develop predictive models that identified patterns between C. neoformans and antifungal drugs, as well as orthologous genes from Aspergillus fumigatus identified via OrthoFinder. Results: The Gradient Boosting Classifier demonstrated the highest accuracy at 82%, with an AUC of 70% and a Matthews Correlation Coefficient of 0.71. The study also identified 276 synergistic drug combinations, including those involving 2-aminobenzothiazole, which showed potential when combined with FK506, 5-fluorocytosine, and 4-hydroxytamoxifene. Conclusion: The MOSCAF model successfully integrated chemogenomics, orthology mapping, drug structural information, and machine learning to identify synergistic drug combinations for Cryptococcus neoformans and Aspergillus fumigatus. The model achieved an AUC of 0.81, effectively predicting combinations involving 2-aminobenzothiazole, demonstrating its potential in accelerating antifungal drug combination discovery and addressing drug-resistant fungal infections through bioinformatics-driven approaches.