Re: [Rdkit-discuss] Canonical smiles for medium and large rings?

[Greg L. <gre...@gm...>]

James,

On Tue, Jan 4, 2011 at 6:29 PM, James Davidson <J.D...@ve...> wrote:
>
>> It would be *really* useful to have some more real-world
>> cases like this one to use as tests. So if you happen to have
>> others you can send I would be quite happy to have them.
>
> On that note, I have added a comment to the bug tracker
> (https://sourceforge.net/tracker/?func=detail&aid=3139534&group_id=16013
> 9&atid=814650) - but was not sure how to attach a file (eg sdf) there,
> so apologies for it ending up on more lines than I intended...  Also, I
> logged in with my google account, but it looks like it may not be clear
> who it is!

Thanks for these. I just added a couple of initial tests based on
them. I will try to find the time to make them a bit more
comprehensive in the next couple of days.

> The first two examples are two marine natural products that only differ
> in the geometry of the double bond in the medium ring.  The final
> example is a cis- analogue that I synthesised during my PhD for which a
> crystal structure was also obtained.  The stereochemistry in these
> systems is 'challenging' to say the least, so I thought they would make
> reasonable test cases.  I should say that even for the cis- double bond
> cases, RDKit does a rather ugly job of the 2D depiction - but I am not
> sure if other depictors will perform much better...

Yeah, I'm afraid it's not going to do  a reasonable job with the
depiction of natural products. Most depictors (including many human
ones) have trouble getting these rendered well.

> On a related note, I was keen to manually double-check the
> stereochemistry that had been assigned to each of the chiral centres
> (particularly the ones involving the 9-5 ring connections - as these are
> potentially troublesome), and found myself wishing there was a way to
> easily label a 2D depiction of the molecules with the atom ID.  What I
> ended-up doing was the following:
>
> 1.  Getting the R/S info + atomIdx back from RDKit (example output):
>>>> Chem.FindMolChiralCenters(mol)
> [(3, 'R'), (7, 'R'), (8, 'S'), (9, 'R'), (11, 'R'), (18, 'R'), (24,
> 'R')]
> 2.  Opening the molfile in a program where I know how to label with atom
> IDs (pymol)
> 3.  Check which atom is which manually (had to add 1 to the RDKit
> atomIdx values as they start at 0) then double-check with reference
> values.
>
> RDKit performed admirably - but I presume this is dependant on the
> quality of the wedge info coming in from the SDF(?)

If the data are read from an SDF, yes: the initial stereochem
information comes from the SDF. If you have a 3D SD file, you can also
have the RDKit ignore bond wedging and assign chirality based purely
on coordinates.
R/S assignments are done in a later step; it's always nice to hear
that those are correct.

for what it's worth: I tend to use Marvin Sketch for the "drawing
molecules with atom indices to check up on stereochemistry" task. It
will also assign absolute stereochem to atoms and bonds (usually
correctly), so it's a useful check there too.

Best regards,
-greg