SETE

SETE predicts T cell receptor (TCR)–epitope binding specificity by applying sequence-based ensemble learning to CDR3β-derived features for improved epitope recognition.

Key Features:

• Ensemble learning (gradient boosting decision tree): SETE employs a gradient boosting decision tree ensemble to analyze patterns within TCR sequences.
• CDR3β k-mer decomposition: SETE deconstructs Complementarity-Determining Region 3 beta (CDR3β) sequences into short adjacent-amino-acid chains (k-mers) as model features.
• Sequence-based feature design: SETE focuses on sequence-derived features and reduces reliance on time-consuming feature extraction from V(D)J gene loci and on biophysical characteristics of amino acid molecules.
• Performance on VDJdb: Reported experimental results indicate SETE outperforms existing state-of-the-art methods in predicting TCR epitope specificity when evaluated on the VDJdb dataset.
• Multi-classification capability: SETE is formulated to perform multi-class classification of TCR–epitope specificity.

Scientific Applications:

• Immunotherapy development: Predicting TCR–epitope interactions to inform therapeutic vaccine and cancer-treatment research.
• Understanding TCR repertoires: Providing insight into how k-mers formed by adjacent amino acids influence epitope recognition within TCR repertoires.

Methodology:

SETE decomposes CDR3β sequences into short adjacent-amino-acid chains (k-mers), uses sequence-derived features instead of V(D)J-locus or biophysical-feature extraction, applies an ensemble learning strategy implemented as a gradient boosting decision tree, and is evaluated as a multi-class classification model on the VDJdb dataset.