TABLE 1.

Dataset statistics. The “Interaction” column means the unique count of pairs of {CDR3α, CDR3β, peptide}, and CDR3αβ denotes the unique count of pairs of {CDR3α, CDR3β}. The duplication count, the “in duplication” row of the “Unique count” column, means the number of unique data that are shared between training and test sets, i.e., overlapped data count. The “Pos. rate” column denotes the positive ratio in the binary label.

Dataset	Unique count	CDR3αβ	Peptide	Interaction	Pos. rate
McPAS	In training	3,181	316	23,363	0.1665
McPAS	In test	833	190	4,729	0.1512
-	In duplication b/w training and test	132	171	0	N/A
VDJdb-without10x	In training	2,902	175	19,526	0.1670
VDJdb-without10x	In test	689	120	4,010	0.1504
-	In duplication b/w training and test	111	111	0	N/A
Combined data dataset (A)	In training	23,299	478	119,046	0.1400
Recent data test set (B)	In test	33,183	838	33,360	0.1667
COVID-19 dataset (C)	In test	1,676	1265	2,120,140	1.887 ⋅ 10–5
-	In duplication b/w (A) and (B)	18	44	0	N/A
-	In duplication b/w (A) and (C)	1	0	0	N/A