[Dock-fans] Is the following a reasonable application for DOCK? (Filtering/screening scheme)

John J. Irwin jji at cgl.ucsf.edu
Thu Sep 27 09:39:23 PDT 2007


Hi Mike

Thanks for your question about docking to many targets to select ligands
that bind well to all or most of them. Although I have not done exactly
what you are describing, I have some opinions that I am willing to share ;-)

I know exactly what you mean about DOCK attempting and succeeding to
generate a pose for most ligands, which seems to work against the
protocol as you have described it.

One modification you might try is to dock the entire database to all
targets, and then look for consensus - possibly score based, but perhaps
safer rank based. How you compute an aggregate score is also up in the
air. One approach might be to just sum the rank each molecule gets in
each run - so like golf, smaller is better. Another might be to sum
int(log(rank)), thus a molecule could score well against 9 but poorly
against 1 and not be penalized too heavily for it. A third way might be
to just count 1 or 0 if the molecule is in the top 100 or 1000 or
whatever, but this risks being arbitrary.

Please bear in mind that some ligands do erroneously well or poorly for
often arcane reasons (insufficient sampling, a bug) as well as good
reasons (the ligand just doesn't fit as well, etc).

I hope this is some help.

Good luck

John


Silberstein, Michael wrote:
> Hi all,
>
>  
>
> I have a question regarding a particular docking application, and I would like to get a sense as to whether DOCK is a
>
>  reasonable program for implementing such an application, or if other docking programs might be better served.
>
>  
>
> I would like to perform virtual screening (using a mini-library of approximately 100,000 compounds) on a number of target
>
>  crystal structures (approximately ten structures in all).  The architecture of the substrate-binding pocket of each of these
>
>  structures is generally quite similar; however, small sub-regions of the pocket are, nevertheless, different/unique in order to
>
>  accomodate specificity differences amongst the various ligands/substrates that will bind to each individual protein.
>
>  
>
> I am aiming to find "common inhibitors" that will be capable of binding ALL of the structures with reasonable binding
>
>  affinities (rather than to find separate inhibitors for each target protein).  The protocol by which I was hoping to identify such
>
>  "common inhibitors" is the following:
>
>  
>
>  
>
> Step 1:  Dock all 100,000 to target structure #1 using DOCK.  Based on DOCK's bump filter and scoring function, only a
>
>  certain subset of the 100,000 compounds would provide docking output, while the remainder (that might have too many
>
>  bump clashes, etc.) would be skipped.
>
>  
>
> Step 2:  Dock the remaining (unskipped) compounds into target structure #2.  As in Step 1, some molecules will be
>
>  skipped and others will be retained.
>
>  
>
> Step 3:  Dock the remaining compounds (i.e. the ones retained after Steps #1 and #2) into target structure #3.  Some
>
>  molecules are further retained.  Others are not.
>
> ...
>
> ...
>
> ...
>
>  
>
> Step N:  Dock the compounds (that have been retained after docking to the (N-1) targets) to the final target structure.  Only 
>
> those remaining compounds that are retained after this final docking step can be used in subsequent studies in order to 
>
> further narrow down this subset and identify the inhibitors common to each of the (N) proteins.
>
>  
>
>  
>
>  
>
> My question is whether or not DOCK can be used to implement the above protocol and to filter out (after each step) the
>
>  molecules that do not appear to dock well from the remaining molecules in the initial mini-library?  Or might an alternative
>
>  docking program be better served for this application and protocol?
>
>  
>
> Thus far in my usage of DOCK (both v4.0.1 and v6.1), I have found that the VAST majority ( >99%) of compounds have been
>
>  retained (and not skipped), which has made me question the feasibility of performing this "filtering" analysis using DOCK.  
>
>  
>
> Also, is there a way to increase the "stringency" of DOCK (either based on scoring function selection, bump filter, or other
>
>  parameters) so that fewer compounds can be retained from step to step? 
>
>  
>
> If anyone is able to address these particular issues/questions, I would greatly appreciate it.  Thanks in advance.
>
>  
>
>  
>
> Sincerely,
>
> Mike Silberstein
>
>  
>
>  
>
>  
>
> _______________________________________________
> Dock-fans mailing list
> Dock-fans at docking.org
> http://blur.compbio.ucsf.edu/mailman/listinfo/dock-fans
>   


More information about the Dock-fans mailing list