Llama-TCR: Generate De Novo TCR with Large Language Model

T cell receptor-engineered T cell (TCR-T), where T cells are equipped with engineered antigen-specific receptors, is a promising approach of cancer immunotherapy. However, it is known that naturally occurred TCR is not able to detect some cancer antigens due to negative selection and resulted in the proliferation of cancer cells. This study introduces a novel application of large language models (LLM) in the de novo generation of T cell receptors (TCR). We propose Llama-TCR, a generative model trained on database of antigen and TCR pairs, for diverse and functional candidate TCR sequence generation. Using both sequence based benchmarking and 3D structure inspection, we show that the model is able to generate TCR sequences targeting to given antigens and has great potential in the development of novel TCR-T immunotherapy.