Skip to main content

Table 5 Comparisons of performance with state-of-the-art methods on regression datasets, splitting the datasets by scaffold splitting in a ratio of 8:1:1 (lower values are better)

From: Double-head transformer neural network for molecular property prediction

Methods

Lipophilicity

PDBbind

MolNet [1]

\(0.655\pm 0.036\)

\(1.920\pm 0.070\)

RF on Morgan [19]

\(0.908\pm 0.052\)

\(2.011\pm 0.240\)

FFN on Morgan [19]

\(1.045\pm 0.042\)

\(2.737\pm 0.518\)

FFN on Morgan Counts [19]

\(1.003\pm 0.068\)

\(3.015\pm 0.636\)

FFN on RDKit [19]

\(0.792\pm 0.032\)

\(1.842\pm 0.252\)

DMPNN [19]

\(0.648\pm 0.057\)

\(1.858\pm 0.300\)

Ours

\({{ {0.590}}}\pm {{ {0.038}}}\)

\({{{1.599}}}\pm {{ {0.199}}}\)