Title: New Benchmarking Reveals Weak Generalization in TCR Antigenic Epitope Prediction Models

Abstract: While precise computational forecasting of T cell receptor (TCR) antigen specificity holds the potential to revolutionize T cell biology research and facilitate scalable immune engineering, current models suffer from inadequate sensitivity and specificity for widespread use. A significant bottleneck in this field is the lack of rigorously defined, unseen benchmark datasets, which are essential for conducting unbiased evaluations of model performance and generalizability. This study introduces two complementary types of datasets that satisfy this requirement, positing that they establish a sturdy framework for assessing models and serve as a basis for developing next-generation TCR-antigen prediction algorithms.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC