ADL: Automating the Autodock

mswingle at jaguar1.usouthal.edu mswingle at jaguar1.usouthal.edu
Sat Apr 4 00:35:44 PDT 2009


----- Original Message -----
From: Jack Shultz <jshultz at hydrogenathome.org>
Date: Friday, April 3, 2009 1:13 pm
Subject: Re: ADL: Automating the Autodock
To: autodock at scripps.edu, Riad Schulz <rschulz at uni-greifswald.de>


> Hello,
> 
> I am considering an experimental design that would perform virtual
> screening using flexible receptors. I've already setup a good system
> using the mgltools scripts. Now I am considering the flexible
> receptors. When you prepare a flexible receptor you need to specify
> which residues should be flexible. I am wondering if there are any
> approaches to picking these receptors that would lend themselves to
> automation.
> 
> Jack

Jack,

If you just mean automating the preparation when the flexible residues are user-supplied, 
there is a python script that comes with ADT that does that. Most likely, though, you're 
looking for a way to automate the identification of which residues should be made flexible. 
I don't think that there are any clear-cut methods for doing this unless you have a variety 
of different experimental structures available for the receptor in question or you're willing 
to run MD simulations. 

There have been attempts to predict flexible regions from the amino acid sequence (see 
e.g. http://cubic.bioc.columbia.edu/services/profbval/ ). I'm not that familiar with the 
literature but, IIRC, most of these studies use B-factors as a proxy for flexibility, which 
IMHO is a somewhat dubious assumption to make except for higher resolution structures. 
The other thing to consider is that we are not so much interested in residues that are 
inherently "jittery" in a given structure but rather those which may switch to a 
conformation that differs from that in the available structure. It is certainly possible 
(maybe even probable) that a residue involved in ligand binding could take on different 
highly stable (i.e. presumably low B-factor) conformations when different ligands are 
bound to the receptor. Thus, a low B-factor observed in one structure does not necessarily 
mean that the residue wouldn't switch to a new conformation in a different complex. 
OTOH a high B-factor might imply a predilection for changing conformation that would be 
useful to model during docking. 

If I were going to use B-factors as selection criterion, I think it would probably be best to 
take them from structures of unliganded receptors at ~2.5A resolution or better (B's can 
soak up a lot of model errors during refinement and should be viewed skeptically in low 
resolution structures). Also, they should probably be normalized in some fashion since B's 
are influenced by the resolution, the refinement protocols, and the crystal packing. It 
might also be useful to try to fit a TLS (translation libration screw-axis) model to the B's to 
try to separate out large-scale anisotropic motions of the whole receptor or domains from 
local motions (see e.g. http://skuld.bmsc.washington.edu/~tlsmd/ ).

Other considerations would include the type of residue (lys, arg, glu e.g. are notorious) and 
it's environment (solvent accessibility, h-bonding etc.).

Hope this helps.

Regards,

Mark


More information about the autodock mailing list