NetTCR-2.1

NetTCR-2.1 predicts binding probabilities between T-cell receptor (TCR) complementarity-determining region (CDR) loops and major histocompatibility complex class I (MHC-I) peptides to model TCR specificity for peptide-MHC complexes.

Key Features:

• Predictive Modeling Framework: Employs a modeling framework that addresses data redundancy and defines modeling objectives to handle high variability in TCR–pMHC data.
• Integration of CDR Loops: Integrates information from all six CDR loops (CDR1, 2, and 3) of the TCR to improve prediction specificity.
• Peptide-Specific Modeling: Utilizes peptide-specific models rather than pan-specific models to focus on specific TCR–MHC-peptide interactions and improve predictive performance.
• Negative Data Construction: Constructs negative datasets using a combination of true negatives and mislabeled negatives to refine discrimination between binding and non-binding events.
• Machine Learning Integration: Incorporates machine learning models that outperform similarity-based approaches, particularly as distance from training data increases.
• Validation Strategy: Recommends validating predictive power using similarity-based modeling approaches and evaluating performance as a function of distance to training data.

Scientific Applications:

• High-throughput TCR identification: Supports high-throughput identification of peptide-specific TCRs from sequence data.
• Assay development and mechanistic studies: Aids development of assays and studies that elucidate the rules governing TCR–pMHC interactions.

Methodology:

Computational methods include comparison of peptide-specific versus pan-specific models, integration of all six CDR loops, construction of negative datasets combining true negatives and mislabeled negatives, handling of data redundancy, application of machine learning models versus similarity-based approaches, and performance evaluation as a function of distance to training data.