IFNepitope2

IFNepitope2 predicts and designs interferon-gamma (IFN-γ) inducing peptides to identify and engineer epitopes for immunological research and vaccine development.

Key Features:

• Dataset: Uses 25,492 experimentally validated human IFN-γ inducing peptides and 7,983 experimentally validated mouse IFN-γ inducing peptides.
• Machine Learning Integration: Applies machine learning classification models trained on peptide features.
• Feature Encoding: Emphasizes compositional features, notably dipeptide composition, and compares encodings such as one-hot and binary profiles.
• Model Performance: Extra tree classifiers achieved AUROC of 0.89 for human peptides and 0.83 for mouse peptides.
• Hybrid Model: Combines the best-performing machine learning model with BLAST similarity search to yield improved AUROC of 0.90 for humans and 0.85 for mice.
• Independent Validation: Models were evaluated on independent datasets not used in training or testing and outperformed existing techniques on those validations.

Scientific Applications:

• Vaccine Design: Enables design of subunit and peptide-based vaccines by predicting IFN-γ inducing epitopes.
• Host-specific Immune Annotation: Supports annotation and analysis of host-specific IFN-γ inducing peptides in humans and mice.

Methodology:

Developed classification models using machine learning on peptide compositional features including dipeptide composition with comparisons to one-hot and binary encodings; selected extra tree classifiers as best-performing; implemented a hybrid approach combining the ML model with BLAST; and evaluated models on independent datasets.