A Graph Feature Auto-Encoder for the prediction of unobserved node features on biological networks

BMC Bioinformatics

Table 3 The average MSE of predicting test expression values \(\tilde{\mathbf {X}}_{test}\) using different models

Method	TF_Net		PPI		Genetics
Method	Features	Graph	Features	Graph	Features	Graph
GCN	7.791 ± 3.550	15.127 ± 2.280	6.208 ± 0.607	11.106 ± 0.52198	5.988 ± 0.696	4.560 ± 0.351
GraphSAGE	0.332 ± 0.160	8.078 ± 2.592	0.265 ± 0.135	2.844 ± 0.349	0.233 ± 0.127	4.466 ± 1.605
GraphConv	0.318 ± 0.154	13.812 ± 4.534	0.308 ± 0.139	3.094 ± 0.431	0.234 ± 0.116	5.226 ± 1.054
FeatGraphConv	0.285 ± 0.135	7.525 ± 2.941	0.244 ± 0.130	5.207 ± 1.476	0.201 ± 0.112	3.414 ± 0.691
LR-embedding	1.583 ± 0.200	2.279 ± 0.403	1.091 ± 0.166	1.453 ± 0.264	1.863 ± 0.332	1.653 ± 0.271
RF-embedding	1.945 ± 0.318	2.150 ± 0.363	1.472 ± 0.267	1.452 ± 0.267	1.883 ± 0.343	1.897 ± 0.351
MLP	0.424 ± 0.170	–	0.354 ± 0.153	–	0.332 ± 0.134	–
LR	0.215 ± 0.126	–	0.147 ± 0.105	–	0.108 ± 0.084	–
RF	0.507 ± 0.143	–	0.194 ± 0.103	–	1.882 ± 0.343	–

(Bold indicates lowest error mean per category of experiments for each group of methods (end-to-end, indirect, or baseline)

ISSN: 1471-2105