Table 3.

Model utilities by the configurations and the training datasets used.

Dataset	BIOSNAP (BS)	DAVIS (DV)	BindingDB (BDB)	Integration (INT)
Non-pretrained (NP) ($\varphi$ + $\varphi$)	NP-Model-BS	NP-Model-DV	NP-Model-BDB	NP-Model-INT
ChemBERTa-pretrained (CP) (ChemBERTa + $\varphi$)	CP-Model-BS	CP-Model-DV	CP-Model-BDB	CP-Model-INT
ProtBERT-pretrained (PP) ($\varphi$ + ProtBERT)	PP-Model-BS	PP-Model-DV	PP-Model-BDB	PP-Model-INT
Full-pretrained (FP) (ChemBERTa + ProtBERT)	FP-Model-BS	FP-Model-DV	FP-Model-BDB	FP-Model-INT

Note: ϕ indicates the model that did not employ the pretrained weights of the ChemBERTa and ProtBert models. Integration refers to a model trained by merging the three datasets.