User 21b7e0228c
22-03-2009 07:43:11
Hi there,
Despised by medicinal chemists, nitro groups were typically the beasts on hand of which one explained the need for structure standardization. Since I do not have any discovery collaborations with AlQaida, nor with James Bond (I just read somewhere that in the latest movie NITROMETHANE is... the newest superexplosive - how stupid can Holywood get??), I paid little attention to these otherwise honorable groups.
Now, for historic reasons (I'm still haunted by the shadow of my organic chemistry teacher "don't expect to pass if you draw five-legged nitrogens in nitro!"), we chose the standard nitro representation with split (formal) charges, by contrast to the above-incriminated ChemAxon-preferred five-legged N version. At that point, I had checked, from the sketcher interface, that the logP/logD calculation is, indeed, independent on the representation - and it is... as soon as you stick to Tools=>Partition=>logwhatever.
Last week, I however realized (with horror!) that this ain't true for generatemd. Take, for example, TNT:
echo "CC1=C(C=C(C=C1[N]([=O])=O)[N]([=O])=O)[N]([=O])=O" | generatemd c -k LogP -D returns a decent 2.31
echo "CC1=C(C=C(C=C1[N+]([O-])=O)[N+]([O-])=O)[N+]([O-])=O" |generatemd c -k LogP -D returns a nasty -8.84, which is a bit overdone (even for a Schwarzenegger movie).
It's less blatant with fewer nitro groups, but.. I'm now in doubt about all the QSAR models I fitted using logP as a descriptor (note - if they fitted well, with the wrong logPs, then they should be applicable with the wrong logPs as well!). Luckily I don't have so many nitros to deal with... but perhaps other groups are concerned, if generatemd confuses FORMAL and REAL charges...
Now, I don't know whether I'm doing' something wrong or whether this punishment has been inflicted on me by the Vatican, for unorthodox [uncatholic] comments - but shouldn't generatemd take care all alone of internal standardizations that are essential for property predictions?
Cheers!
Dragos
Despised by medicinal chemists, nitro groups were typically the beasts on hand of which one explained the need for structure standardization. Since I do not have any discovery collaborations with AlQaida, nor with James Bond (I just read somewhere that in the latest movie NITROMETHANE is... the newest superexplosive - how stupid can Holywood get??), I paid little attention to these otherwise honorable groups.
Now, for historic reasons (I'm still haunted by the shadow of my organic chemistry teacher "don't expect to pass if you draw five-legged nitrogens in nitro!"), we chose the standard nitro representation with split (formal) charges, by contrast to the above-incriminated ChemAxon-preferred five-legged N version. At that point, I had checked, from the sketcher interface, that the logP/logD calculation is, indeed, independent on the representation - and it is... as soon as you stick to Tools=>Partition=>logwhatever.
Last week, I however realized (with horror!) that this ain't true for generatemd. Take, for example, TNT:
echo "CC1=C(C=C(C=C1[N]([=O])=O)[N]([=O])=O)[N]([=O])=O" | generatemd c -k LogP -D returns a decent 2.31
echo "CC1=C(C=C(C=C1[N+]([O-])=O)[N+]([O-])=O)[N+]([O-])=O" |generatemd c -k LogP -D returns a nasty -8.84, which is a bit overdone (even for a Schwarzenegger movie).
It's less blatant with fewer nitro groups, but.. I'm now in doubt about all the QSAR models I fitted using logP as a descriptor (note - if they fitted well, with the wrong logPs, then they should be applicable with the wrong logPs as well!). Luckily I don't have so many nitros to deal with... but perhaps other groups are concerned, if generatemd confuses FORMAL and REAL charges...
Now, I don't know whether I'm doing' something wrong or whether this punishment has been inflicted on me by the Vatican, for unorthodox [uncatholic] comments - but shouldn't generatemd take care all alone of internal standardizations that are essential for property predictions?
Cheers!
Dragos