Extended Data Fig. 8: Performance of the retrained CDR3β-only models on low prevalence of true TCR-epitope pairs.
From: Assessment of computational methods in predicting TCR–epitope binding recognition
a-c, Performance of CDR3β-only models using AS negatives in predicting seen-epitope test data (a), independent test (b) and unseen-epitope independent test (c) data with different prevalences (0.1%, 1%, 10%, and 50%) of positive samples in terms of Precision, F1, Recall and Specificity. In consideration of the relatively small magnitude of many metric values, all metrics were rounded to three decimals to enable clearer comparison of subtle performance differences across models.
