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

Table 4 shows the results of the ablation experiments

Model	Node-central MPN	Edge-central MPN	Side-chain embedding	Side-chain adding	Gaussian mixture distribution	LeDock Success rate	GraphDTA Success rate
1		√	√	√	√	0.458 ± 0.206	0.536 ± 0.299
2	√		√	√	√	0.656 ± 0.240	0.604 ± 0.339
3	√	√		√	√	0.565 ± 0.378	0.721 ± 0.350
4	√	√	√		√	0.259 ± 0.231	0.250 ± 0.248
5	√	√	√			0.014 ± 0.043	0.019 ± 0.070
6	√	√	√	√	√	0.720 ± 0.326	0.776 ± 0.333

For each metric, the best result among all baseline models is represented as bold format

ISSN: 1758-2946