TABLE 3.

Docking mean scores comparison between original and de-strained poses with standard error.

	Model	Affinity (Vina) $\downarrow$	CNN affinity $\uparrow$	CNN VS score $\uparrow$
Original Poses	Random	$-5.08\pm 0.01$	$6.70\pm 0.00$	$3.63\pm 0.0$
	Pocket2Mol	$-\mathbf{11.73}\mathbf{\pm }\mathbf{0.03}$	$8.01\pm 0.01$	$6.66\pm 0.02$
	DiffSBDD	$-10.83\pm 0.02$	$\mathbf{8.55}\mathbf{\pm }\mathbf{0.02}$	$6.66\pm 0.01$
	PharmacoForge	$-10.39\pm 0.01$	$8.40\pm 0.12$	$\mathbf{8.30}\mathbf{\pm }\mathbf{0.09}$
Destrain Poses	Pocket2Mol	$-10.59\pm 0.03$	$7.71\pm 0.02$	$3.68\pm 0.04$
	DiffSBDD	$-7.65\pm 0.03$	$6.80\pm 0.03$	$4.06\pm 0.03$
	PharmacoForge	$-\mathbf{10.49}\mathbf{\pm }\mathbf{0.02}$	$\mathbf{7.73}\mathbf{\pm }\mathbf{0.02}$	$\mathbf{4.98}\mathbf{\pm }\mathbf{0.04}$

Best score for each column is listed in bold.