Table 2.
Impact of individual groups of descriptors on random forest prediction of bond dissociation energies a
| No. of descriptors | RMSD | MAD | MaxError | |
|---|---|---|---|---|
| Selection 2 b |
293 |
7.01 |
4.35 |
56.87 |
| Selection 2 - CN point |
209 |
9.06 |
5.50 |
80.97 |
| Selection 2 - Element pair |
218 |
7.18 |
4.48 |
57.13 |
| Selection 2 - Fragment point |
288 |
7.02 |
4.36 |
58.75 |
| Selection 2 - Aromatic fragment point |
279 |
7.07 |
4.39 |
62.14 |
| Selection 2 - In-ring fragment point |
281 |
7.03 |
4.36 |
55.31 |
| Selection 2 - No-type pair |
269 |
7.06 |
4.43 |
58.95 |
| Selection 2 - No-type bond-breaking difference pair |
284 |
6.96 |
4.32 |
57.70 |
| Selection 2 - π fragment point |
281 |
7.03 |
4.37 |
58.59 |
| Selection 2 - Molecular element pair |
263 |
6.96 |
4.31 |
57.86 |
| Selection 2 - Molecular CN fragment point |
265 |
7.04 |
4.35 |
57.77 |
| Selection 3 c | 209 | 7.00 | 4.32 | 58.36 |
a Results are in kcal/mol, and were obtained in the out-of-Bag (OOB) RF validation procedure over the training set.
b Combination of the following descriptors: 1) Connection number point descriptors, 2) Simple element pair descriptors, 3) Fragment point descriptors (only the field with the lower value is used), 4) Aromatic fragment point descriptors with corresponding molecular descriptors, 5) In-ring fragment point descriptors, 6) No-type pair descriptors, 7) No-type bond-breaking difference pair descriptors, 8) Conjugated π system fragment point descriptors with corresponding molecular descriptors, 9) Simple element molecular pair descriptors, 10) Molecular connection number fragment point descriptors. Descriptors 1, 4, 5, are calculated for spheres from 0 to 5. Descriptors 2, 3, 6, 8 are calculated for spheres from 1 to 5. Descriptor 2 only involves pairs at a distance of one bond. Descriptor 7 involves pairs with interatomic distances from 2 to 7. Descriptor 9 involves pairs with interatomic distances of 1–2.
c Combination of Descriptors 1, 2, 4, 6, 8.