On Thu, Jul 31, 2008 at 3:43 PM, Rajarshi Guha <[EMAIL PROTECTED]> wrote: > Reproducible is definitely a worthy model. But equally important is > quality.
Surely we aim at the most statistically sound *and* most predictive model. But anyone can do that. > You're right that given the small training set, it might a bit tricky > to get a good quality model. But are you thinking of straight up > QSAR? Not generally... I am not aware of polymorph prediction tools that are open source, but I am not saying that we should not take that approach... actually... > Recent papers [1,2] suggests that crystal lattice energies > should be taken into account - but for that they used structures from > the CCDC. I fully agree... the solubility is certainly not just a function of the molecular structure, but of the (polymorph) crystal structure too... Indeed, it's an interesting challenge :) Polymorph prediction in itself already is a challenge (there is a biannual(?) competition for that), and correlating those polymorphs to solubility is a whole other story... force fields for crystal structures have their limitations... So, a simple QSAR approach would be rather quick-and-dirty, but might not do that bad... Though, if we have the expertise *and* tools within the Blue Obelisk community to go the polymorph route, I'd most welcome that. In which case I would actually be able to use some of my PhD study results, where I developed a vectorial descriptor for crystal structures :) Is there anyone listening in who if there are open source tools that allow polymorph prediction? > At the very least, we'd need some sort of solvent enabled > conformer generator. Coming up with reasonable conformers is an important step in polymorph prediction, and actually make PP extra hard to do. > I don't know whether taking into account > multiple confs (say via Boltzmann weighting) would be useful Also > good partial charges would be useful (MOPAC would be a good option > for that). SVN is back up again. So, partial charges would be a good way to contribute to this effort. We could use those for regular QSAR descriptors too. > [1] http://dx.doi.org/10.1021/mp7000878 Yeah, might simply be a 1-2 for the UCC :) Though I in a good competion people from the organizing institute(s) are not allowed to participate, so not sure John Mitchell will :) Peter, are you actually allowed to join in, with CML as standard for data exchange in this endavour? > It might also be useful to derive some sort of interaction > descriptors from MD runs (described in [2]) > > [2] http://dx.doi.org/10.1021/mp070030+ Have not read [2]... it is using open source tools? Egon -- ---- http://chem-bla-ics.blogspot.com/ ------------------------------------------------------------------------- This SF.Net email is sponsored by the Moblin Your Move Developer's challenge Build the coolest Linux based applications with Moblin SDK & win great prizes Grand prize is a trip for two to an Open Source event anywhere in the world http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Blueobelisk-discuss mailing list Blueobelisk-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/blueobelisk-discuss
