[Open Babel] Antwort: Re: Serializing OBMol to memory

[<ern...@ba...>]

I already use fingerprints. :-)

The problem arises afterwards:

Let's say a substruct search for 'c1ccccc1F' yields 1810 hits from the 
screening stage. 1805 are valid, 5 are false positives. In order to find 
those false positives, all candidates have to be checked again through the 
SMARTSPattern.

The main performance killer there is the instantiation of each target 
molecule from a textual representation.

By changing from SMILES to V2000 molfile alone, I gained about 30% more 
throughput speed (because V2000 has explicit bonds I suppose). I further 
suppose, that saving a OBMol representation that already has ringset, 
aromaticity and whatnot perceived at registration time and just reading 
this back into memory should be even faster.

Best Regards

Ernst-Georg Schmid