CTLPred

CTLPred predicts cytotoxic T lymphocyte (CTL) epitopes directly from antigenic sequences to support subunit vaccine design and studies of T cell-mediated immunity.

Key Features:

• Direct prediction approach: Employs a direct method that leverages patterns inherent in T cell epitopes rather than relying solely on MHC class I binding affinities.
• Machine learning techniques: Integrates Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Quantitative Matrix (QM) methods trained on 1,137 experimentally validated MHC class I-restricted T cell epitopes.
• Performance metrics: Reports QM at 70.0% accuracy, ANN at 72.2%, and SVM at 75.2% accuracy evaluated by Leave One Out Cross-Validation (LOOCV), balancing sensitivity and specificity.
• Consensus prediction: Combines outputs of multiple machine-learning methods to produce consensus predictions that outperform QM-based predictions on blind datasets.
• Subgroup analysis: Distinguishes true T-cell epitopes from MHC binders (non-epitopes) via subgroup analysis.
• MHC restriction prediction: Predicts MHC restriction for predicted CTL epitopes.

Scientific Applications:

• Subunit vaccine development: Identifies CTL epitopes to inform selection and design of subunit vaccine candidates.
• Immunological and bioinformatics research: Supports immunologists and bioinformaticians investigating T cell-mediated immunity and epitope identification.

Methodology:

Support Vector Machines (SVM), Artificial Neural Networks (ANN) and Quantitative Matrix (QM) methods were trained and tested on a dataset of 1,137 experimentally validated MHC class I-restricted T cell epitopes and evaluated using Leave One Out Cross-Validation (LOOCV); consensus predictions combine machine-learning outputs and subgroup analysis contrasts true T-cell epitopes with MHC binders.