A cell-based kinetic framework enables TCR specificity prediction.

The ability to predict T-cell receptor (TCR) specificity from sequences could transform immunotherapy, vaccine development, and our understanding of immune recognition. However, progress has been shaped by "edge cases", in which specificity appears to be captured by simplified descriptors, such as sequence motifs or correlations between binding affinity and functional activation. Although informative, these regimes are not representative of the general mode of TCR recognition. Emphasizing such cases has contributed to a drift in the field, where both experimental assay design and computational modeling increasingly rely on nonrepresentative signals, limiting generalizability across antigens and TCRs. We argue that this drift stems from a lack of a clear biophysical definition of TCR specificity and continued reliance on equilibrium binding assays that are not well suited to capture it. These limitations propagate into training datasets, constraining the performance and generalizability of predictive models. To address them, we introduce three key elements. First, we develop a cell-based assay for quantitative measurement of TCR-pMHC binding kinetics in a physiological context. Second, we introduce a mechanistic framework for interpreting these data, showing that the widely used reversible ligand-receptor model is insufficient in the general case and proposing the TCR cycle model as a minimal systems-level description. Third, we describe a strategy for generating multiplexed, high-throughput datasets and integrating mechanistic modeling with machine learning. Together, this work establishes a foundation for a mechanistically grounded and scalable approach to TCR specificity prediction.