Fig. 5: Performance of retrained models across sample sizes.
From: Assessment of computational methods in predicting TCR–epitope binding recognition
a, Performance of models trained with different TCR counts per epitope in seen-epitope prediction. Dot plots show per-antigen AUPRC, with adjacent heatmaps showing overall AUPRC. b, AUPRC comparison of models (n = 24) from a across TCR count groups. Colored dots represent individual model AUPRC, black dots indicate the mean and error bars represent the mean ± s.d. c, Correlation between epitope-associated TCR counts and AUPRC for the top three models from a of the >300 group. Dots represent epitopes, colored by antigen group. P values were from two-sided t-tests (n = 53). d, Performance saturation analysis on the top three models using five epitopes with the greatest TCR counts, showing per-epitope AUPRC and mean performance (red line).
