SwarmTCR

SwarmTCR predicts T cell receptor (TCR) specificity by combining nearest-neighbor classification with an optimized, weighted CDR-distance metric to infer antigen recognition from TCR sequence data.

Key Features:

• Nearest-Neighbor Classification: Uses labeled TCR sequence data with a nearest-neighbor approach to assign specificity based on sequence distances.
• Particle Swarm Optimization (PSO): Applies particle swarm optimization to tune weights assigned to each complementary determining region (CDR) loop for improved classification.
• Weighted CDR-Distance Metric (modified TCRdist): Computes distances by aligning CDR loop sequences with the BLOSUM62 matrix and applying per-CDR weights to alignment scores.
• Biologically Interpretable Weights: Produces optimized weights for CDR regions that can be interpreted in terms of their relative contribution to specificity.
• Compatibility with Data Types and Chains: Processes both single-cell and deep (bulk) sequencing data and supports analysis of α and β TCR chains.

Scientific Applications:

• Epitope-specific TCR classification: Demonstrated improved classification accuracy for epitopes including EBV-BRLF1300, EBV-BRLF1109, NS4B214–222 (single-cell) and EBV-BRLF1300, EBV-BRLF1109, IAV-M158 (bulk α and β chains).
• Method benchmarking: Shows superior performance compared to TCRdist in comparative studies of TCR specificity prediction.
• Immune-response studies: Applicable to analyses of immune responses in contexts such as infectious diseases, cancers, autoimmunity, cancer immunotherapy, and immunopathology.

Methodology:

Distance between TCRs is computed using a modified TCRdist by aligning CDR loop sequences with BLOSUM62, multiplying alignment values by optimized per-CDR weights found via particle swarm optimization, summing weighted values to obtain the distance, and applying a nearest-neighbor classifier on labeled sequence data.