Figure 1.

Flow chart shows the basic procedure for preparing different datasets. After collecting data from public resources and merging the preprocessed into one dataset (TCR preprocessed dataset, tpp dataset), different filtering criteria were applied to obtain the positive examples for $d_{base,strict}$ , $d_{base,uniform}$ , $d_{bal}$ and $d_{imbal}$ datasets. Negative examples were generated within folds (refer to 4.1.3) after splitting (refer to 4.1.2) to obtain the complete datasets. $d_{base}$ : the base dataset filtered from tpp dataset. $d_{base,strict}$ : strict splitting used on $d_{base}$ . $d_{base,uniform}$ : uniform splitting used on $d_{base}$ . $d_{bal}$ : the balanced dataset filtered from $d_{base}$ , then split using uniform splitting. $d_{imbal}$ : the imbalance dataset filtered from $d_{base}$ , then split using uniform splitting.