Repertoire-level generation of T-cell epitopes with a large-scale generative transformer
Bibliographic record
Abstract
Single-cell TCR sequencing enables high-resolution analysis of T Cell Receptor (TCR) diversity and clonality, offering valuable insights into immune responses and disease mechanisms. However, identifying cognate epitopes for individual TCRs requires complex and costly functional assays. We address this challenge with EpitopeGen, a large-scale transformer model based on the GPT-2 architecture that generates potential cognate epitope sequences directly from TCR sequences. To overcome the scarcity of TCR-epitope binding pairs (≈ 100,000), EpitopeGen uses a semi-supervised learning method, termed BINDSEARCH, which searches over 70 billion potential pairs and incorporates high binding affinity predictions as pseudo-labels. To incorporate CD8 + T cell biology into the model as an inductive bias, EpitopeGen employs a novel data balancing method, termed Antigen Category Filter, that carefully controls antigen category ratios in its training dataset. EpitopeGen significantly outperforms baseline approaches, generating epitopes with high binding affinity, diversity, naturalness, and biophysical stability. Notably, the epitopes generated by EpitopeGen follow biologically plausible antigen category distributions, a crucial feature not achieved by other models. Using EpitopeGen, we directly identify subsets of clonally expanded tumor-infiltrating lymphocytes that recognize tumor-associated antigens, exhibiting elevated cytotoxicity and reduced exhaustion markers. From COVID-19 patients, EpitopeGen detects T cells that recognize COVID-19 spike proteins and non-structural proteins with distinct transcriptomic characteristics. In conclusion, EpitopeGen represents the first computational method that enables direct inference of antigen recognition profiles of CD8 + T cells from plain TCR repertoires.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".