Table 3.
Comparison of accuracy value between SeBioGraph and other link prediction methods on five biomedical graph datasets.
| Method | CTD CDA | NDF-RT DDA | DrugBank DDI | STRING PPI | STITCH CPI |
|---|---|---|---|---|---|
| Matrix factorization | |||||
| SVD (De Lathauwer et al., 2000) | 93.6 ± 0.2% | 77.9 ± 0.3% | 91.9 ± 0.1% | 86.7 ± 0.1% | 31.7 ± 0.4% |
| LLE (Roweis and Saul, 2000) | 86.5 ± 0.3% | 89.7 ± 0.4% | 89.1 ± 0.2% | 79.8 ± 1.0% | 29.4 ± 0.3% |
| LE (Belkin and Niyogi, 2003) | 85.6 ± 0.4% | 93.0 ± 0.3% | 79.6 ± 0.2% | 63.9 ± 2.1% | 23.2 ± 0.5% |
| GF (Ahmed et al., 2013) | 88.4 ± 0.4% | 72.0 ± 0.6% | 88.2 ± 0.3% | 81.7 ± 0.5% | 32.1 ± 0.3% |
| GraRep (Cao et al., 2015) | 96.0 ± 0.1% | 96.3 ± 0.1% | 92.5 ± 0.1% | 89.4 ± 0.1% | 41.4 ± 0.4% |
| HOPE (Ou et al., 2016) | 95.1 ± 0.1% | 94.9 ± 0.1% | 92.3 ± 0.1% | 83.9 ± 0.1% | 42.7 ± 0.2% |
| Random walk | |||||
| DeepWalk (Perozzi et al., 2014) | 92.9 ± 0.2% | 78.3 ± 0.4% | 92.1 ± 0.1% | 88.4 ± 0.1% | 26.4 ± 0.3% |
| Node2vec (Grover and Leskovec, 2016) | 91.1 ± 0.2% | 81.9 ± 0.5% | 90.2 ± 0.1% | 82.8 ± 0.3% | 37.7 ± 0.6% |
| Struc2vec (Ribeiro et al., 2017) | 96.5 ± 0.1% | 95.8 ± 0.1% | 90.4 ± 0.1% | 90.9 ± 0.1% | 44.0 ± 0.1% |
| Graph Neural networks | |||||
| LINE (Tang et al., 2015) | 96.5 ± 0.1% | 96.2 ± 0.2% | 90.5 ± 0.2% | 85.9 ± 0.3% | 36.2 ± 0.4% |
| GAE (Tang et al., 2016) | 93.7 ± 0.1% | 81.3 ± 0.7% | 91.7 ± 0.1% | 90.0 ± 0.1% | 35.8 ± 0.1% |
| SDNE (Wang et al., 2016) | 93.5 ± 1.0% | 94.4 ± 0.4% | 91.1 ± 0.6% | 88.4 ± 0.8% | 37.8 ± 0.8% |
| Our model | |||||
| SeBioGraph | 97.2 ± 0.5% | 96.4 ± 0.6% | 93.1 ± 0.3% | 89.9 ± 0.6% | 48.8 ± 0.7% |
| - Auxiliary | 93.8 ± 0.5% | 87.1 ± 0.5% | 88.9 ± 0.3% | 85.6 ± 0.4% | 39.1 ± 0.5% |