Journal: bioRxiv
Article Title: TAPIR: a T-cell receptor language model for predicting rare and novel targets
doi: 10.1101/2023.09.12.557285
Figure Lengend Snippet: ( a ) We extended TAPIR with a generative component capable of producing TCR sequences specific to target antigens. We then queried the extended model to sample and score 10,000 paired TCR alpha and beta CDR3 sequences against HLA-A2 presented influenza antigen peptide GILGFVFTL. Six TCR designs with the highest TAPIR scores were constructed and transduced into the TCR negative Jurkat NFAT-reporter line. Each TCR was evaluated on its ability to bind to HLA-A2-GILGFVFTL tetramer complex and to elicit T-cell activation signals (NFAT) when stimulated by GILGFVFTL-presenting antigen presenting cells (T2 cells). ( b ) The generative extension to TAPIR learns to predict CDR3 regions for alpha and beta TCRs given a target of interest and a starting set of V and J genes. First the alpha CDR3 region is predicted given the target and V and J genes, then the beta CDR3 region is predicted given the alpha chain, target sequence, and V and J gene set. Following TCR sequence generation, the full sequence is scored with the downstream original TAPIR model ( c ) TAPIR scores, normalized binding MFI, and T-cell activation results of six AI-designed TCRs against A2-presented GILGFVFTL. Tetramer binding MFI (median fluorescence intensity) was normalized to negative control TCR tetramer binding MFI. T-cell activation NFAT positive gate was set based on the top 1% of NFAT values in the negative control TCR group. Three AITCRs showed statistically significant binding to the target antigen (n=3, * p<0.05, ** p<0.001). Two AITCRs activated Jurkat cells when interacting with GILGFVFTL-presenting cells (n=3, ** p<0.001). ( d ) Correlation between predicted key amino acids with modeled TCR-antigen interactions for AITCR-5. The co-crystal structure of AITCR-5 and A2-presented GILGFVFTL antigen was modeled by AlphaFold 2 (red = alpha CDR3, blue = blue CDR3, orange = antigen peptide, green = HLA-A2, yellow dash line = predicted hydrogen bond). Amino acid importance scores for the binding regions of the CDR3s were computed with TAPIR. We highlight key amino acids (high scores) involved in the TCR-antigen and TCR alpha-beta hydrogen bond interactions (yellow dashed lines).
Article Snippet: Wild type Jurkat cells (TIB-152), K562 cells (CCL-243), HEK293T cells (CRL-3216), and T2 (CRL-1992) cells were obtained from the ATCC.
Techniques: Construct, Activation Assay, Sequencing, Binding Assay, Fluorescence, Negative Control