VDJdb, McPAS-TCR (2020, December) and IEDB (2021, December) databases were used for creation of training sets.

The training sets include 37191 unique CDR3 sequences of alpha chain TCR for 894 epitopes (116 epitopes were selected for prediction) and 32505 unique CDR3 sequences of alpha chain of TCR for 86 MHC types (25 MHC types were selected for prediction) and 144409 unique CDR3 sequences of beta chain of TCR for 1377 epitopes (202 epitopes were selected for prediction) and 50153 unique CDR3 sequences of beta chain of TCR for 99 MHC types (28 MHC types were selected for prediction).

Only epitopes and MHC types with IAP (equivalent of AUC) calculated by LOO CV procedure more 0.75 were selected for TCR-Pred. The mean IAP values calculated by LOO CV procedure were 0.874 for prediction of epitopes for alpha chain of CDR3 TCR, 0.860 for prediction of MHC types for alpha chain of CDR3 TCR, 0.883 for prediction of epitopes for beta chain of CDR3 TCR and 0.886 for prediction of MHC types for beta chain of CDR3 TCR.