Re: [Cdk-devel] atom typing Pubchem

[Christoph S. <er...@do...>]

Rajashi,

this is *extremely* helpful for future development.
Thanks a lot!
What I would be interested in is: How long did it take to run the test
(on what machine)?

It would be great to get this running as a "nightly" and seeing the
types of atoms which fail flagged up in color in 2D diagrams.

Cheers,

Chris

Rajarshi Guha wrote:
> Hi, I've put in a QA project that will be running the new atom types  
> against Pubchem
> 
> As part of the testing, I ran 50,000 molecules through it and here's  
> a summary
> 
> No. of molecules exhibiting type errors = 5060
> No. of elements involved in type errors = 100
> No. of molecules with bad C types       = 2699
> No. of molecules with bad N types       = 0
> No. of molecules with bad O types       = 1889
> No. of molecules with bad P types       = 396
> No. of molecules with bad S types       = 186
> 
> The no of molecules and elements is unique.
> 
> Also, the numbers for individual elements are overestimates. This is  
> because for a number of molecules such as 139358 [1], 14694528 [2]  
> and 485468 [3] the structures are inorganic complexes - in these  
> cases all the C's or O's are flagged as untyped. Not surprising!
> 
> So the fact that less than 10% of this sample flags missing atom  
> types is not too bad.
> 
> One way to work on this is to avoid using the whole of Pubchem but  
> work on something like a drug-like subset. That way we'd avoid  
> inorganic complexes. However, the issue becomes - where do we store  
> it? I'd suggest storing a compressed version in the cdk-qa repo
> 
> [1] http://pubchem.ncbi.nlm.nih.gov/summary/summary.cgi? 
> db=pccompound&term=139358
> [2] http://pubchem.ncbi.nlm.nih.gov/summary/summary.cgi?cid=14694528
> [3] http://pubchem.ncbi.nlm.nih.gov/summary/summary.cgi? 
> db=pccompound&term=485468
> 
> -------------------------------------------------------------------
> Rajarshi Guha  <rg...@in...>
> GPG Fingerprint: 0CCA 8EE2 2EEB 25E2 AB04  06F7 1BB9 E634 9B87 56EE
> -------------------------------------------------------------------
> I saw Elvis.  He sat between me and Bigfoot on the UFO.
> 
> 
> 
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2005.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Cdk-devel mailing list
> Cdk...@li...
> https://lists.sourceforge.net/lists/listinfo/cdk-devel


-- 
PD Dr. Christoph Steinbeck
Lecturer in Chemoinformatics
Univ. Tuebingen, WSI-RA, Sand 1, D-72076 Tuebingen, Germany
Phone: (+49/0) 7071-29-78978   Fax: (+49/0) 7071-29-5091

What is man but that lofty spirit - that sense of enterprise.
... Kirk, "I, Mudd," stardate 4513.3..