DLpTCR

DLpTCR predicts interactions between T cell receptors (TCRs) and peptides presented by major histocompatibility complex (MHC) molecules using a multimodal ensemble of deep-learning models to identify immunogenic T cell epitopes.

Key Features:

• Multimodal ensemble: Integrates three deep-learning models into a single multimodal ensemble to improve predictive accuracy and robustness.
• TCR-peptide interaction prediction: Predicts likelihoods of interactions between TCRs and antigenic peptides presented by MHC molecules.
• Single and paired chain support: Predicts interactions for both single TCR chains and paired TCR chains.
• Evaluation on independent datasets: Validated on independent datasets including COVID-19 data and IEDB data with reported performance metrics.
• Reported performance: Achieved area under the curve (AUC) up to 0.91 for single-chain predictions and overall accuracy of 81.03% for paired-chain predictions on IEDB data.
• Generalization of interaction rules: Demonstrates the ability to learn general interaction rules and generalize across different antigen peptide recognitions by TCRs.

Scientific Applications:

• Immunogenic epitope identification: Identification of immunogenic T cell epitopes for downstream experimental validation.
• Vaccine development: Prioritization of peptide candidates for vaccine design based on predicted TCR recognition.
• Cancer immunotherapy: Selection of candidate neoantigens or peptide targets for TCR-based cancer immunotherapies.
• SARS-CoV-2 / COVID-19 immune analysis: Analysis and prediction of TCR responses to SARS-CoV-2 peptides using COVID-19 datasets.

Methodology:

Integrates three deep-learning models into a multimodal ensemble to predict interactions between single and paired TCR chains and peptides and was evaluated on independent datasets including COVID-19 and IEDB, reporting AUC up to 0.91 and 81.03% accuracy for paired chains.