Hi, On 10 December 2010 15:13, Thomas Strunz <beginn...@hotmail.de> wrote:
> Hi again, > > first a short update if anyone is interested: > > Not very surprisingly Java serialization is quite useless. I tried it > (HSQLDB has a special column type for this) but to keep it simple, it is > considerably slower than just reading the Molfiles. > This is an interesting finding. > Not to mentioned that the serialized form seems to use a large amount of > disk space. > > So back to storing properties. How do you store aromaticty? Just Yes/no per > molecule? > We don't store per-molecule aromaticity flags, but do store atom and bond aromaticity instead. > Isn't that covered by the fingerprint? > Indirectly it is covered. > (an hence useless) > It's useful for the subsequent graph matching. > > As far as i understood, each Atom and Bond that are part of an aromatic > system have the according flag set. > > This is probably relevant in graph-matching and hence when creating a > molecule, aromaticity must be detected. > Yes. > So every aromatic Atom and bond per molecule would have to be stored in the > database so that this calculation is not needed anymore. Of course that > would be a stupid idea to do so. > That's exactly what we do :) > Ideally it should be in the Molfile but it's not. > > Creating AtomContainers seems to be pretty straight forward, not much to > improve. Maybe a light-weight implementation only for structure search would > be an option? > I would say making atom containers lighter should be possible, but will require some efforts to do profiling and find where optimization is possible. IMHO the most time consuming part is aromaticity detection / ring finding and avoiding that by storing pre-calculated flags is a useful workaround. Regards, Nina > > Reagrds, > > Thomas > >
------------------------------------------------------------------------------
_______________________________________________ Cdk-user mailing list Cdk-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/cdk-user
