On 10 May 2010 01:06, Leonid Chepelev <leonid.chepe...@gmail.com> wrote: >> Just to note, comp chem does a local optimisation. To do a global >> optimisation you need to do a conformer search and locally optimise >> each. With OB, what you're actually comparing is the accuracy of >> MMFF94 geometries and those from B3LYP. This is already well studied. >> > > Yes, I know that this is well studied, but I am aiming to see whether > OpenBabel does as well as CDK relative to a "sort of true" structure taken > from a reference database and reoptimized with DFT in gas phase (and > consequently, whether I should prefer one to the other for a particular set > of cases), not whether MMFF does a comparable job to DFT in geometry > optimization.
I understand - just checking. > Obviously if one of the toolkits shows a result that is > considerably different from the other, two things could be happening: A) > there were bugs in implementing the 0D to 3D conversion or B) one toolkit > has a slightly more accurate/exhaustive geometry optimization technique > (very hard to imagine). If A is the case, I will see large individual errors > in geometries, and if B is the case, I will observe a consistent small > difference. My primary concern is case A - that is, whether the > three-dimensional geometries are properly assigned, given handling chiral > centers and various other points where 3D geometry generation can go wrong. > If you have seen a comparison between the two (or more) toolkits along these > lines or any sort of benchmarking studies, I would be very grateful if you > could point me to them. Of course, an easier thing to do here is to try the > "broken telephone" test - that is, whether all the canonical SMILES I feed > OpenBabel's 3D generator will be the same as those obtained after converting > the generated 3D geometries back into canonical SMILES... Let us know how it goes. I don't know of any such comparison between toolkits. I know that in the development version stereochemistry is handling correctly (except for ring templates with stereochemistry), but you are likely to find some problems in the release. > Speaking of conformational searches, I had been under the impression that > some sort of rudimentary conformational analysis, as has been implemented in > obconformer, was already taking place when I called gen3D - but from what > you have said I am guessing that this is not the case and I would have to > hack something together to enable this. I've just checked the source (http://openbabel.svn.sourceforge.net/viewvc/openbabel/openbabel/branches/openbabel-2-2-x/src/ops/gen3d.cpp?revision=3339&view=markup) and it seems that it should be okay. The lines are as follows: pFF->SteepestDescent(250, 1.0e-4); pFF->WeightedRotorSearch(200, 25); pFF->ConjugateGradients(250, 1.0e-6); The weighted rotor search is a type of conformer search. > In any case, thank you very much for your helpful comments! ------------------------------------------------------------------------------ _______________________________________________ OpenBabel-discuss mailing list OpenBabel-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/openbabel-discuss
