Table 1.

Performance of $\mathrm{\Delta }$-AEScore compared to the ${\mathrm{\Delta }}_{\text{vina}}$ for affinity prediction on the CASF-2013 and CASF-2016 benchmarks. For $\mathrm{\Delta }$-AEScore the “hard overlap” between the training and both test sets is removed while for ${\mathrm{\Delta }}_{\text{vina}}$ only the “hard overlap” between the training set and CASF-2013 is removed [49, 67]. The best performance for each test set is underlined. RMSE values are given in pK units

Model	Training set	Test set	RMSE	Pearson’s r
$\mathrm{\Delta }$-AEScore${}^{†}$	Refined 2013	CASF-2013	1.53	0.74
$\mathrm{\Delta }$-AEScore${}^{†}$ (no H)	Refined 2013	CASF-2013	1.52	0.74
${\mathrm{\Delta }}_{\text{vina}}$ [49]	Refined 2013	CASF 2013	—	0.69
Vina (optim)	—	CASF-2013	1.82	0.61
$\mathrm{\Delta }$-AEScore${}^{†}$	Refined 2016	CASF-2016	1.34	0.79
$\mathrm{\Delta }$-AEScore${}^{†}$ (no H)	Refined 2016	CASF-2016	1.32	0.80
${\mathrm{\Delta }}_{\text{vina}}$ [40, 49]	Refined 2013	CASF 2016	—	0.81
Vina (optim)	—	CASF-2016	1.75	0.59

${}^{†}$ This work