Table 11 Average test set MSE of the best performing MAT models for each representation and dataset
From: Extended study on atomic featurization in graph neural networks for molecular property prediction
Representation | Rat \(\downarrow\) | Human \(\downarrow\) | QM9 \(\downarrow\) | ESOL (random) \(\downarrow\) | ESOL (scaffold) \(\downarrow\) |
|---|---|---|---|---|---|
F A | 0.205 0.325 | 0.241 0.260 | 501.171 582.430 | 0.135 0.246 | 0.214 0.483 |
A + N A + H A + C A + R A + A | 0.235 0.231 0.317 0.233 0.237 | 0.258 0.231 0.269 0.255 0.253 | 569.508 559.458 583.693 517.893 570.120 | 0.184 0.269 0.233 0.203 0.242 | 0.309 0.290 0.415 0.351 0.287 |
F-N F-H F-C F-R F-A | 0.216 0.201 0.204 0.202 0.207 | 0.240 0.254 0.240 0.241 0.239 | 497.775 486.866 497.051 544.970 488.815 | 0.201 0.161 0.128 0.200 0.136 | 0.207 0.243 0.221 0.371 0.229 |