Re: [Cdk-jchempaint] Behaviour with adding detached fragments

[Egon W. <ego...@gm...>]

On Wed, Mar 4, 2009 at 4:58 PM, Stefan Kuhn <ste...@eb...> wrote:
> On Wednesday 04 March 2009 15:43:40 Egon Willighagen wrote:
>> On Tue, Mar 3, 2009 at 6:16 PM, Stefan Kuhn <ste...@eb...> wrote:
>> > I found JCP right now adds a fragment
>>
>> What kind of fragments are you taking about?
> Any sort of thing in JCP, e. g. a ring, a chain, whatever. You can call it
> molecule or atomcontainer or whatever, I don't care

But it matters... I understand you mean not "t-Bu", but just an
IAtomContainer with IAtoms and IBonds...

>> > to an existing AtomContainer, when you
>> > paint e. g. a ring detached from the existing structure. This behaviour
>> > causes problems with creating reactions.
>>
>> What are those problems?
> If you draw two separate fragments, molecules, atomcontainers or whatever you
> call them and want to make one the reactant and one the product (which is the
> natural way to do it in jcp) and you click on the first and say "make
> reactant in new reaction", both will become one reactant, since make
> reactant/product operates on atomcontainers.

So, there is a bug in that code: the make reaction action should
clearly do a partitioning... it has to convert the model backend
anyway... that does not seem to happen right now.

Clearly, this functionality has not been ported properly from JCP2 to JCP3.

>> > Obviously the reaction stuff can be
>> > changed, but I would prefer to have fragments in the MoleculeSet
>> > separatly.
>>
>> How would you need to change the reaction stuff?
> Do a partitioning

Sure. See above.

>> > What else have we got the MoleculeSet for?
>>
>> A fragment is not a molecule... So, MoleculeSet is not for a set of
>> fragments, I'd say.
> In jcp, this is all interchangeable. What defines a molecule in JCP? Does it
> suddenly become a molecule if the user finishes? This distinction is not
> viable in jcp

No, but from your writing I understood that you wanted for making that
distinction...

>> > Plus in some cases (e. g. the
>> > reaction stuff) you would need to do a partitioning, which is costly.
>>
>> How so? And why is this a problem? Upon deleting a bond you would
>> still have to do this, ...?
>>
>> > But in any case we need a procedure to follow. I vote for one
>> > AtomContainer must always be a connected structure. What do you think?
>>
>> IMolecule has the (unenforced) semantics that it has to be connected.
>> Such requirement has never existed for IAtomContainer.
> And chemmodel contains a MoleculeSet. So right now you can create illegal
> molecules.

Yes, that would be a correct observation.

>> Please explain the problems in more detail...
> I hope I made it clear.

The current CDK design has some limitations, and if you convert a JCP
drawing of one or more molecules into a reaction, you will have to do
partitioning. If you like the performance drop because of that or not.

Does that answer your questions?

Egon

-- 
Post-doc @ Uppsala University
http://chem-bla-ics.blogspot.com/