Table 4 Comparisons of performance with state-of-the-art methods on classification datasets, splitting the datasets by random splitting in a ratio of 8:1:1 (higher values are better)
From: Double-head transformer neural network for molecular property prediction
Methods | PCBA | BACE | Tox21 | SIDER |
|---|---|---|---|---|
MolNet [1] | \(0.136 \pm 0.004\) | / | \(0.829 \pm 0.006\) | \(0.648 \pm 0.009\) |
RF on Morgan [19] | / | \(0.825 \pm 0.039\) | \(0.619 \pm 0.015\) | \(0.572 \pm 0.007\) |
FFN on Morgan [19] | \(0.263 \pm 0.008\) | \(0.873 \pm 0.040\) | \(0.788 \pm 0.017\) | \(0.652 \pm 0.010\) |
FFN on Morgan Counts [19] | \(0.268 \pm 0.006\) | \(0.882 \pm 0.030\) | \(0.790 \pm 0.020\) | \(0.638 \pm 0.020\) |
FFN on RDKit [19] | \(0.207 \pm 0.005\) | \(0.858 \pm 0.034\) | \(0.832 \pm 0.016\) | \(0.654 \pm 0.019\) |
DMPNN [19] | \(0.769 \pm 0.010\) | \(0.892 \pm 0.031\) | \(0.839 \pm 0.022\) | \(0.657 \pm 0.016\) |
Ours | \({{ {0.821}}} \pm {{ {0.005}}}\) | \({{{0.923}}} \pm {{ {0.035}}}\) | \({{ {0.847}}} \pm {{{0.015}}}\) | \({{ {0.679}}} \pm {{ {0.015}}}\) |