TABLE 4.

The performance of AttnTAP under varied TCR feature extraction models.

		ACC a	REC	PRE	F1	AUC
McPAS -TCR	I b	0.736	0.803	0.708	0.752	0.827
	II	0.762	0.803	0.743	0.772	0.854
	III	0.766	0.807	0.747	0.775	0.857
	IV	0.774	0.755	0.755	0.758	0.861
	V	0.781	0.818	0.762	0.789	0.869
VDJdb	I	0.840	0.820	0.855	0.837	0.906
	II	0.839	0.799	0.868	0.832	0.901
	III	0.842	0.806	0.869	0.836	0.904
	IV	0.844	0.820	0.861	0.840	0.908
	V	0.847	0.829	0.870	0.844	0.914

^aAbbreviations: ACC: accuracy; REC: recall; PRE: precision; F1: F1 score; AUC: area under the receiver operating characteristic curve.

^bModel numbers: I: the multilayer perceptron model; II: the two-layer long short term memory (LSTM) model; III: the one-layer bi-directional LSTM model; IV: the two-layer LSTM model with attention mechanism; and V: Attn-BiLSTM model.