TCRfinder is a deep learning pipeline designed for sequence-based screenings of T-cell receptors (TCRs) and neoantigens. The core strategy involves developing accurate interaction scoring models using the sequences of the β-chain TCR CDR3 region and target peptides. The process begins with the pre-training of two separate language models for TCR and peptide sequences. These pre-trained models are then combined to form a joint embedder model from the concatenated representations of the TCR and peptide language models. Transformer blocks are applied to both paired and individual TCR and peptide representations, creating new embedding matrices. Finally, a Multi-Layer Perceptron (MLP) block layer is used to develop a TCR-peptide interaction scoring model based on these new embeddings. To tackle the challenges of TCR and neoantigen recognition, TCRfinder trains two distinct models: one for peptide-based TCR screening and another for TCR-based neoantigen screening. Check [Help] page for more details. Please report any problems and questions on our [Discussion Board].
zhanglab
zhanggroup.org
| +65-6601-1241 | Computing 1, 13 Computing Drive, Singapore 117417