How to decompose the UFF (or MMFF94) scoring of a small molecule? #6377
Replies: 4 comments 2 replies
-
@ptosco maybe? |
Beta Was this translation helpful? Give feedback.
-
Ok, good to know. |
Beta Was this translation helpful? Give feedback.
-
Single point energies on force fields are very fast. As I mentioned on the e-mail list, this is unlikely to be much faster. You're forgetting that if you move a dihedral, you also have to re-calculate all the non-bonded terms. There are usually far more of those (approx. n^2) than bonds and angles. For MMFF94, you'd need to update both electrostatic and VdW terms. At that point, you're doing 90-95% of the single point evaluation anyway. If you're evaluating lots of geometries and lots of molecules, the best approach is to generate multiple geometries and use multi-threading over the geometries and run molecules in batches / parallel. (Speaking as someone who's run ~3M molecules / 150M conformers recently.) |
Beta Was this translation helpful? Give feedback.
-
If this becomes necessary, I could decompose a ligand into rigid groups (they are separated by RotBonds). |
Beta Was this translation helpful? Give feedback.
-
I have a ligand, I would like to score its current conformer using rdkit's UFF implementation.
Later, I would like to change some rotatable bond (single bond out of ring) and update
the conformer's energy bu just re-evaluating the part of the energy that is supposed to have
changed (i.e. the dihedral component).
Bond lengths, bond angles and partial charges being constant.
I suspect it should be faster than rescoring the conformer from scratch.
How to do this with rdkit?
Thanks a lot,
F.
Beta Was this translation helpful? Give feedback.
All reactions