[Zinc-fans] requesting peer suggestions on docking and duplication

[John J. Irwin]

Hi Martin

That's interesting. We like to keep database tranches small so we can 
run dock in parallel (course grained parallel, completely separate 
processes) on dozens or even hundreds of cores, and then combine the 
results together at the end. Most computers nowadays come with 4 or even 
8 effective cores inside, so running dock in smaller chunks makes 
efficient use of your available hardware.

John
UCSF ZINC Team

Martin Stoermer wrote:
> Mornign everyone,
>
> another thing to remember is that if you do concatenate the zinc files 
> (as we do) is that some docking packages will run into memory problems 
> after too many ligands. For example we have had problems with GOLD. 
> When I concatenated all 113 of the ZINC Druglike sdf "chunks" you get 
> a 9.6GB SD file. When I ran a GOLD job on our cluster across 8 nodes 
> if dies after ~600K ligands with error messages like:
>
> libpvm [t40002]: pvm_pklong(): Value too large
> Pvm Function pvm_pklong( (long) start_point, 1, 1), called from 
> GOLD_COMM_SendNextDock, caused an error: Value too large
>
> ******************************************************************************
> warning: Slave process 786433 on host node21.***.***.***.*** failed 
> with following error message:
> Gold internal failure (3.2) in docking ligand 0 in file 
> /chem_db/Drug_like/GOLD/Drug_like_3d.sdf
> ******************************************************************************
> WARNING: pvm task problem 786433
>  Fatal error: 
> Received interrupt [signal number 15]
>
> I can't speak for how Autodock, Dock and the others handle these large 
> datasets, I haven't tried to bludgeon them with 2.5 million compounds 
> yet. Maybe others will have that answer. So it may be more sensible to 
> limit yourself to smaller sets e.g. one concatentated vendor database 
> at a time.
>
> cheers,
> Martin
>
>
>>  
>> The question is, if we merge all the files into one single giga SDF 
>> file, are the ZINC IDs unique to all the entries, meaning, can we 
>> safely traceback the ligand of interest to its representative pH subset?
>>  
>>  
>> With best regards,
>> Elite158
>>  
>> _______________________________________________
>> Zinc-fans mailing list
>> Zinc-fans at docking.org <mailto:Zinc-fans at docking.org>
>> http://blur.compbio.ucsf.edu/mailman/listinfo/zinc-fans
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Zinc-fans mailing list
> Zinc-fans at docking.org
> http://blur.compbio.ucsf.edu/mailman/listinfo/zinc-fans
>