Table 3.

Comparison to previous methods.^a

	TPP2 AUROC	TPP2 AP	TPP3 AUROC	TPP3 AP
EPIC-TRACE (our) [${\mathcal{D}}_{\alpha \beta ,\alpha ,\beta }$]	0.906 $\pm$ 0.000	0.691 $\pm$ 0.001	0.693 $\pm$ 0.008	0.294 $\pm$ 0.007
EPIC-TRACE (our)	0.906 $\pm$ 0.000	0.698 $\pm$ 0.000	0.691 $\pm$ 0.008	0.291 $\pm$ 0.005
ERGO-II	0.895 $\pm$ 0.002	0.659 $\pm$ 0.007	0.675 $\pm$ 0.007	0.274 $\pm$ 0.004
epiTCR	0.793 $\pm$ 0.000	0.581 $\pm$ 0.000	0.515 $\pm$ 0.001	0.183 $\pm$ 0.001
TITAN	0.786	0.454	0.577	0.204
ImRex	0.697	0.420	0.519	0.178

^aEPIC-TRACE, ERGO-II, and epiTCR are evaluated on five 10-fold cross-validation runs, whereas TITAN and ImRex are evaluated on only one of the five cross-validations due to long training time. Reported values are the mean of the five 10-fold cross-validation runs together with the standard error. Values for best performing models are bolded.