Re: [Rdkit-discuss] molecule standardization in cartridge search

[Tim D. <tdu...@gm...>]

Jan and others
Thanks for all your suggestions. Just what I need.

Tim

On 26/09/2015 12:26, Jan Holst Jensen wrote:
> Hi Tim,
>
> Soren (cc:ed) wrote me and asked about molvs. Thanks to Soren for 
> reminding that the original question was about standardization more 
> than calling Python code from Postgres :-).
> http://molvs.readthedocs.org/en/latest/
>
> Take a look at molvs - it's got lots of functionality that you will 
> need. We also use molvs as the backbone for much of our standardization.
>
> Cheers
> -- Jan
> ________________________
>
> Hi Tim,
>
> A simple getting-started example is:
>
>     CREATE FUNCTION smiles2molfile(smiles text) RETURNS text
>       LANGUAGE plpythonu AS $$
>     import rdkit
>     from rdkit import Chem
>
>     mol = Chem.MolFromSmiles(smiles)
>     return Chem.MolToMolBlock(mol)
>     $$;
>
>
> and you can then
>
>     select smiles2molfile('CC');
>
> and get back a molfile.
>
> For more advanced usage it is worth taking a look at the rdchord 
> project that TJ has sent links to.
>
> Cheers
> -- Jan
>
> On 2015-09-25 15:54, Tim Dudgeon wrote:
>> Jan,
>>
>> thanks for that. I'll give it a try.
>> Are there any examples of writing RDKit functions and procedures for
>> postgres in python?
>> I see this general postgres docs:
>> http://www.postgresql.org/docs/9.4/static/plpython.html
>> but wondered if there are any RDKit specific examples anywhere?
>>
>> Tim
>>
>> On 25/09/2015 08:30, Jan Holst Jensen wrote:
>>> On 2015-09-24 16:22, Tim Dudgeon wrote:
>>>> I'm trying to get to grips with using the RDKit cartridge, and so far
>>>> its going well.
>>>> One thing I'm concerned about is molecule standardization, along the
>>>> lines of the ChemAxon Standardizer that allows substructure searches to
>>>> be done is a way that is largely independent of the quirks of structure
>>>> representation. The classic example would be how nitro groups are
>>>> represented, so that it didn't matter which nitro representation was in
>>>> the query or target structures, because both were converted to a
>>>> canonical form.
>>>>
>>>> My initial thoughts are that this would be done by:
>>>> 1. loading the "raw" structures into a source column that would never be
>>>> changed
>>>> 2. defining a function that performed the necessary transform to
>>>> generate the canonical form of a molecule.
>>>> 3. generating a "canonical" structure column that was the result of
>>>> passing the raw structures through that function
>>>> 4. building the SSS index on that canonical column
>>>> 5. executing queries using that function to canonicalize the query
>>>> structure
>>>>
>>>> The problem I'm finding is that there do not seem to be postgres
>>>> functions defined for doing molecular transforms (essentially a reaction
>>>> transform) and doing things like removing explicit hydrogens. At least
>>>> not in the functions listed on this page:
>>>> http://rdkit.org/docs/Cartridge.html#functions
>>>>
>>>> Am I missing something here, or might I be barking up completely the
>>>> wrong tree?
>>>>
>>>> Tim
>>> Hi Tim,
>>>
>>> We have about the same situation and we're adding standardization
>>> (beyond what RDKit implicitly does when it sanitizes the molecule)
>>> through Python stored procedures. You will need to build and maintain
>>> a normal Python-enabled RDKit installation in parallel to the
>>> cartridge. The Python stored procedures can access the normal RDKit
>>> installation and then run whatever Python code is necessary to do
>>> additional molecule cleanup.
>>>
>>> You will need to tweak your Postgres environment so the Python stored
>>> procedures can load RDKit. This is what I have defined in an
>>> environment file on CentOS:
>>>
>>> RDBASE=/opt/rdkit
>>> LD_LIBRARY_PATH=/opt/rdkit/lib
>>> PYTHONPATH=/opt/rdkit
>>>
>>> On Ubuntu this would go into /etc/postgresql/9.x/main/environment (in
>>> a slightly different format where the values have to be single-quoted).
>>>
>>> Cheers
>>> -- Jan, Biochemfusion
>