Table 2.

Tenfold cross validation performance on TCR split

	ROC-AUC
Model	VDJ	VDJ+COVID-19
K-NN (Baseline)	0.79 ± 0.01	0.78 ± 0.007
AA CDR3	0.75 ± 0.02	0.74 ± 0.007
AA full	0.76 ± 0.007	0.75 ± 0.007
SMI CDR3	0.73 ± 0.007	0.76 ± 0.008
SMI full	0.75 ± 0.006	0.77 ± 0.006
Pretrained	0.81 ± 0.01	0.84 ± 0.005
Pretrained aug.	0.80 ± 0.01	0.86 ± 0.004
Pretrained semifrozen	0.80 ± 0.01	0.84 ± 0.005
Pretrained semifrozen aug.	$0.82\pm 0.01$	$0.87\pm 0.005$

Note: Mean and standard deviation of each model configuration on the VDJ dataset and the VDJ + COVID-19 dataset. Best performance marked in bold.