Distance plus attention for binding affinity prediction

Table 4 Comparison of our method with other state-of-the-art predictors on the CASF-2016 dataset

Predictors	R\((\uparrow )\)	RMSE\((\downarrow )\)	MAE\((\downarrow )\)	SD\((\downarrow )\)	CI\((\uparrow )\)
Pafnucy	0.775	1.418	1.129	1.375	0.789
DeepDTA	0.749	1.443	1.148	1.445	0.771
OnionNet	0.816	1.278	0.984	1.280	0.811
DeepDTAF	0.789	1.355	1.073	1.337	0.799
DLSSAffinity	0.790	1.400	1.135	1.404	0.795
SFCNN	0.792	1.328	1.030	1.331	0.798
\(EGNA^*\)	0.785	1.329	1.105	1.332	0.791
CAPLA	0.843	1.200	0.966	1.202	0.820
DAAP16	0.888	1.061	0.827	1.063	0.853
DAAP20	0.909	0.987	0.745	0.988	0.876
SFCNN (N = 285)	0.793	1.326	1.028	1.325	0.799
EGNA (N = 285)	0.842	1.258	0.980	–	–
ResBiGAAT (N = 285)	0.853	1.230	0.941	–	0.832
DAAP16 (N = 285)	0.886	1.070	0.837	1.071	0.851
DAAP20 (N = 285)	0.908	0.994	0.753	0.996	0.874

\(EGNA^*\) predicted by using our HHM features
N = 285 indicates that the dataset contains 285 protein-ligand complexes, whereas the rest are evaluated with 290 protein-ligand complexes. Missing values are indicated by “-”. The best values are emboldened

ISSN: 1758-2946