Multi-objective de novo drug design with conditional graph generative model

Table 5 Performance of graph based and SMILES based model on property based generation tasks

Condition (C)	\(R_0\)	\(I_0\)	Model	% valid	\(R_C\)	\(EOR_C\)	Diversity (\(I_{\mathbf{c}}\))
\(C_1\)	0.009	0.810	Graph	0.997 ± 0.000***	0.55 ± 0.01***	61	0.798 ± 0.002
\(C_1\)	0.009	0.810	SMILES	0.995 ± 0.001	0.51 ± 0.00	57	0.806 ± 0.000***
\(C_2\)	0.012	0.850	Graph	0.970 ± 0.002***	0.55 ± 0.01**	46	0.841 ± 0.001
\(C_2\)	0.012	0.850	SMILES	0.944 ± 0.001	0.52 ± 0.00	43	0.841 ± 0.001
\(C_3\)	0.011	0.868	Graph	0.957 ± 0.001***	0.35 ± 0.01**	32	0.864 ± 0.001
\(C_3\)	0.011	0.868	SMILES	0.894 ± 0.007	0.31 ± 0.00	28	0.866 ± 0.001**
\(C_4\)	0.008	0.867	Graph	0.929 ± 0.003***	0.73 ± 0.01**	91	0.863 ± 0.001
\(C_4\)	0.008	0.867	SMILES	0.613 ± 0.015	0.66 ± 0.00	82	0.863 ± 0.000

Results are reported as \(Mean \pm SD\). The best performance in each metric is highlighted in italics face. Paired t-tests are carried out for the difference between the graph and SMILES based method (*** for \(p\le 0.001\), ** for \(p\le 0.01\) and * for \(p\le 0.05\))