Anti-EBV

Anti-EBV utilizes artificial intelligence and machine learning to predict potential antiviral compounds against the Epstein-Barr virus (EBV) in lytic phase and identify drug repurposing opportunities.

Key Features:

• AI/ML Predictive Modeling: Implements support vector machine (SVM), k-nearest neighbors (KNN), artificial neural network (ANN), deep neural network (DNN), random forest (RF), and XGBoost algorithms for predictive model development.
• Molecular Descriptor Computation: Computes 17,968 molecular fingerprints and descriptors per molecule, selecting the best-performing 150 descriptors for model input.
• Robustness Validation: Validates models through applicability domain analysis, decoy dataset analysis, and chemical clustering analyses.
• Drug Screening: Screens approved drugs from DrugBank to identify repurposed drug candidates.
• Molecular Docking Validation: Docks top predicted compounds against key EBV lytic proteins BZLF1 and BHRF1 to validate binding affinities.

Scientific Applications:

• Antiviral Drug Discovery: Predicts potential antiviral compounds active against EBV in lytic phase based on IC50/EC50 values from experimental data.
• Drug Repurposing: Identifies approved drugs with potential antiviral activity against EBV, including candidates such as arzoxifene, succimer, abemaciclib, fluspirilene, and suvorexant.
• EBV-Associated Cancer Research: Supports development of therapeutics for EBV-linked malignancies including gastric carcinoma, nasopharyngeal carcinoma, and Burkitt's lymphoma.

Methodology:

Small molecules from the ChEMBL database with experimentally tested antiviral activity against EBV in lytic phase were split into training/testing (T315) and independent validation (V35) datasets. Ten-fold cross-validation was applied, and SVM achieved the best performance with Pearson's correlation coefficient of 0.91 on T315 and 0.95 on V35.