Tricked by Edge Cases: Can Current Approaches Lead to Accurate Prediction of T-Cell Specificity with Machine Learning?

The ability to predict T-cell receptor (TCR) specificity computationally could revolutionize personalized immunotherapies, vaccine development, and the understanding of immunology and autoimmune diseases. While progress depends on obtaining training data that represent the vast range of possible TCR-ligand pairs, systematic assessment of modeling assumptions is equally important and can begin with existing data. We illustrate this by evaluating two ideas currently present in the field: treating TCR specificity and T cell activation as distinct modeling tasks, and using unsupervised models based on sequence similarity for TCR specificity prediction. Although presented as general strategies, we argue these are exceptions rather than universally applicable principles.