iTTCA-RF

iTTCA-RF predicts tumor T cell antigens using machine learning to identify peptides relevant for antitumor vaccine development and molecular function investigations.

Key Features:

• Dataset Utilization: Non-redundant dataset comprising 592 positive tumor T cell antigen samples and 393 negative samples used for model training and validation.
• Feature Encoding Methods: Employs four encoding methods: amino acid composition, global protein sequence descriptors, grouped amino acid composition, and peptide composition.
• Hybrid Feature Representation: Integrates the four encoding methods into a hybrid feature representation to capture diverse peptide characteristics.
• Two-Step Feature Selection: Applies a two-step feature selection procedure that yields 263 informative features from the hybrid set.
• Random Forest Algorithm: Constructs the final predictive model using a random forest algorithm.
• Performance Metrics: Reports balanced accuracy 83.71%, specificity 78.73%, and sensitivity 88.69% in tenfold cross-validation, and balanced accuracy 73.14%, specificity 62.67%, and sensitivity 83.61% on an independent test.

Scientific Applications:

• Cancer Immunotherapy: Supports identification of tumor-specific antigens for development of targeted vaccines and therapies.
• MHC Class I Antigen Prediction: Predicts peptides presented by the major histocompatibility complex class I relevant for CD8+ T cell recognition.
• Molecular Function Investigations: Aids studies that require accurate discrimination between antigenic and non-antigenic peptides.

Methodology:

Uses a non-redundant dataset (592 positives, 393 negatives), four encoding methods (amino acid composition, global protein sequence descriptors, grouped amino acid composition, peptide composition) combined into a hybrid feature set, a two-step feature selection selecting 263 features, a random forest classifier, and evaluation by tenfold cross-validation and an independent test.