TCR-Pred

TCR-Pred predicts interactions between T-cell receptor (TCR) CDR3 sequences and specific epitopes or Major Histocompatibility Complex (MHC) types to identify TCR specificity.

Key Features:

• Predictive Modeling: Represents CDR3 TCR sequences using Multilevel Neighbourhoods of Atoms (MNA) molecular fragment descriptors to build classification models reflecting structure-activity relationships (SAR).
• Algorithmic Approach: Implements a naïve Bayes classifier for probabilistic prediction of TCR–epitope and TCR–MHC interactions.
• Comprehensive Database Integration: Trains models on integrated datasets from VDJdb, McPAS-TCR, and IEDB comprising over 250,000 unique CDR3 TCR sequences.
• Prediction Scope: For α chain CDR3 sequences predicts interactions for 116 epitopes or 25 MHC types, and for β chain CDR3 sequences predicts interactions for 202 epitopes or 28 MHC types.
• Validation and Accuracy: Validates predictive performance with 20-fold cross-validation reporting average AUC values from 0.857 to 0.884.

Scientific Applications:

• T-cell profiling: Assigns probable epitope or MHC specificities to CDR3 TCR sequences for T-cell repertoire profiling.
• Immunotherapy and vaccine research: Supports selection and prioritization of TCR targets in immunotherapy and vaccine development by predicting TCR specificity.
• Mechanistic studies: Enables studies of molecular determinants of T-cell recognition and activation by linking sequence features to epitope and MHC specificity.

Methodology:

CDR3 sequences are encoded with Multilevel Neighbourhoods of Atoms (MNA) descriptors, models are learned using a naïve Bayes classifier on data from VDJdb, McPAS-TCR, and IEDB, and performance is assessed by 20-fold cross-validation yielding AUC values of 0.857–0.884.