Table 4.

ROC-AUC scores on HIV. The best performance is denoted by bold

Tokens	BiGRU	Transformer	TranGRU
-S	0.7927	0.7484	0.8105
-B	0.7942	0.8079	0.8115

“-S” represents using an atom-level tokenizer to tokenize the SMILES strings, and “-B” represents using SPE as the tokenizer