DeepSA: a deep-learning driven predictor of compound synthesis accessibility

Table 1 Performance comparison of the different models on the test set

Model	ACC	Recall	Precision	F–score	AUROC
DeepSA_ChemMTR	0.971	0.968	0.974	0.971	0.997
DeepSA_ChemMLM	0.961	0.955	0.967	0.961	0.995
DeepSA_MinBert	0.939	0.933	0.945	0.939	0.988
DeepSA_TinBert	0.942	0.937	0.947	0.942	0.990
DeepSA_RoBERTa	0.940	0.940	0.940	0.940	0.988
DeepSA_DeBERTa	0.898	0.873	0.920	0.896	0.959
DeepSA_GraphCodeBert	0.938	0.931	0.944	0.937	0.987
DeepSA_SmELECTRA	0.944	0.938	0.949	0.943	0.990

ISSN: 1758-2946