Table 2 Toxicity prediction performances on seven toxicity benchmark datasets
From: Fate-tox: fragment attention transformer for E(3)-equivariant multi-organ toxicity prediction
| Â | BBBP | DILI | Skin Rxn | Carcinogens | SIDER | ClinTox |
|---|---|---|---|---|---|---|
Single-task learning | ||||||
 RF | 67.75 (1.04) | 88.29 (1.98) | 67.71 (2.07) | 75.10 (3.15) | – | – |
 SVM | 68.65 (0.00) | 89.70 (0.00) | 73.12 (0.02) | 78.27 (0.12) | – | – |
 MLP | 63.81 (0.55) | 87.68 (0.50) | 49.93 (13.81) | 78.27 (0.24) | 62.50 (1.11) | 71.95 (1.88) |
 MolCLR | 65.09(0.94) | 81.45 (0.77) | 45.05 (6.17) | 74.00 (3.78) | 59.87 (2.89) | 82.96 (4.24) |
 GraphMVP | 64.24 (1.27) | 89.65 (0.19) | 61.32 (3.50) | 79.51 (4.72) | 61.32 (0.71) | 71.38 (1.49) |
 MAT | 69.08 (4.68) | 89.77 (0.99) | 65.92 (1.13) | 82.99 (3.47) | 62.69 (1.54) | 91.09 (0.41) |
 Molformer | 68.60 (4.64) | 88.98 (0.07) | 64.28 (1.63) | 73.04 (0.21) | 51.41 (0.97) | 71.72 (4.63) |
 Uni-Mol | 68.76 (2.04) | 88.20 (1.69) | 69.48 (4.76) | 82.20 (3.47) | 60.23 (0.91) | 91.11 (3.61) |
 FATE-Tox\(_{\textbf{STL}}\) | 70.15 (1.44) | 90.53 (0.52) | 73.33 (0.61) | 84.16 (2.09) | 63.29 (0.71) | 91.37 (1.53) |
Multi-task learning | ||||||
 FATE-Tox\(_{\textbf{MTL}}\) | 71.16 (1.84) | 91.86 (0.59) | 74.10 (0.84) | 84.78 (0.32) | – | – |