ScaffoldGVAE: scaffold generation and hopping of drug molecules via a variational autoencoder based on multi-view graph neural networks

Hu, Chao; Li, Song; Yang, Chenxing; Chen, Jun; Xiong, Yi; Fan, Guisheng; Liu, Hao; Hong, Liang

doi:10.1186/s13321-023-00766-0

Table 3 The performance comparison between our model and baseline models on scaffold hopping generative model evaluation metrics (SEM) among five distinct targets: CDK2, JAK1, EGFR, LRRK2, and PIM1

From: ScaffoldGVAE: scaffold generation and hopping of drug molecules via a variational autoencoder based on multi-view graph neural networks

Protein	Model	SAscore	GraphDTA				Ledock
Protein	Model	SAscore	Active mean	Active rate	Hop rate	Success rate	Active mean	Active rate	Hop rate	Success rate
CDK2	AAE	2.796	7.559	0.787	0.005	0.007	− 9.522	0.950	0.016	0.012
	VAE	2.934	8.006	0.992	0.003	0.003	− 9.335	0.928	0.014	0.009
	LatentGAN	3.226	7.610	0.863	0.007	0.005	− 9.182	0.882	0.009	0.009
	QBMG	2.860	7.873	0.971	0.003	0.003	− 9.339	0.911	0.010	0.008
	SyntaLinker	2.933	6.753	0.469	0.296	0.113	− 8.164	0.559	0.314	0.176
	REINVENT2	3.230	7.205	0.595	0.128	0.085	− 8.966	0.866	0.027	0.027
	our	3.042	7.151	0.676	1.000	0.676	− 8.208	0.622	1.000	0.622
EGFR	AAE	2.672	8.295	0.929	0.007	0.006	− 10.880	0.947	0.003	0.003
	VAE	2.738	8.304	0.948	0.008	0.008	− 10.960	0.945	0.002	0.002
	LatentGAN	2.866	7.901	0.872	0.003	0.003	− 10.320	0.913	0.002	0.002
	QBMG	2.720	8.221	0.941	0.006	0.006	− 10.980	0.952	0.002	0.001
	SyntaLinker	2.764	6.895	0.506	0.337	0.167	− 8.480	0.721	0.366	0.331
	REINVENT2	2.890	7.187	0.703	0.208	0.156	− 9.505	0.878	0.213	0.212
	our	2.949	7.018	0.613	1.000	0.613	− 8.874	0.923	1.000	0.923
JAK1	AAE	3.106	8.014	0.627	0.001	0.001	− 8.979	0.996	0.001	0.001
	VAE	3.502	8.972	1.000	0.063	0.063	− 8.797	0.996	0.010	0.010
	LatentGAN	4.003	7.984	0.613	0.001	0.001	− 8.778	0.970	0.001	0.001
	SyntaLinker	3.460	7.196	0.283	0.215	0.053	− 7.331	0.498	0.233	0.109
	QBMG	3.439	8.920	1.000	0.053	0.053	− 8.957	1.000	0.012	0.012
	REINVENT2	3.307	6.862	0.300	0.077	0.062	− 8.504	0.871	0.001	0.001
	our	3.510	7.739	0.510	0.952	0.462	− 7.861	0.721	0.955	0.676
LRRK2	AAE	2.682	7.109	1.000	0.062	0.062	− 8.471	0.992	0.064	0.064
	VAE	2.722	7.177	1.000	0.033	0.033	− 7.888	0.991	0.065	0.065
	LatentGAN	2.880	6.981	0.988	0.042	0.042	− 7.878	0.986	0.072	0.072
	SyntaLinker	2.872	6.542	0.860	0.355	0.295	− 6.980	0.839	0.373	0.336
	QBMG	3.396	7.485	1.000	0.072	0.072	− 8.562	0.950	0.028	0.022
	REINVENT2	2.938	6.836	0.973	0.245	0.244	− 8.400	1.000	0.242	0.242
	our	2.915	6.663	0.924	1.000	0.924	− 7.205	0.928	1.000	0.928
PIM1	AAE	2.684	8.179	0.893	0.007	0.007	− 7.869	0.988	0.005	0.005
	VAE	2.803	9.012	0.971	0.013	0.011	− 7.582	0.986	0.016	0.015
	LatentGAN	2.949	8.059	0.885	0.016	0.015	− 7.454	0.956	0.014	0.014
	SyntaLinker	2.938	7.329	0.622	0.239	0.129	− 6.626	0.645	0.242	0.142
	QBMG	2.779	8.648	0.949	0.013	0.012	− 7.661	0.991	0.012	0.012
	REINVENT2	3.095	7.930	0.834	0.193	0.154	− 7.699	0.992	0.172	0.172
	our	3.172	8.310	0.943	1.000	0.943	− 6.879	0.796	1.000	0.796

The best 10% of molecules generated by each model were evaluated. For each metric, the best result among all baseline models is represented as bold format

Back to article page

ISSN: 1758-2946

Contact us

Submission enquiries: journalsubmissions@springernature.com

Journal of Cheminformatics

Contact us