[Dock-fans] Docking to Amber Correlation
William Joseph Allen
william.joseph.allen at gmail.com
Wed Jun 18 20:58:55 PDT 2014
The parameters on that page under the "Virtual Screening" section are okay
except for two. (1) I would recommend to change simplex_grow_max_iterations
to 500. 1000 is fine, but probably just a little bit slower without adding
any benefit. (2) Change num_scored_conformers from 100 to 1. When we do a
virtual screen, we typically only keep the best docked pose for each
compound otherwise it is just too much information to go through.
I apologize for any parameters on these tutorial pages that are misleading
or confusing. These tutorials are written by students who take our
molecular modeling class, and inevitably some things slip through the
cracks. We have not carefully edited the 2014 tutorial yet, so 2013 and
2012 might be more accurate.
To address your question about selectivity, there are many ways to be more
selective when choosing compounds. A general protocol that we have used
successfully in the past looks something like: (1) generate docked poses
using grid score, keep 1 best pose for each molecule. (2) Compute multiple
DOCK scores for each top pose, e.g. continuous energy score AND amber score
AND footprint score. (3) Compute other descriptors using an outside
program. We use MOE to compute things like Lipinski properties, solubility,
logP, # of chiral centers, and others. (4) Create for yourself several
rank-ordered lists. we will make lists of the 1000 top continuous energy
score, the 1000 top footprint score, the 1000 top continuous energy +
footprint score... and several other combinations. (5) Get several people
together, go through each list, look at docked structures one by one,
consider all scores and MOE descriptors, flag things that look promising
for purchase. If your goal is 150 compounds and you have 5 different
rank-ordered list, you may want to choose 30 from each list. Here are a
couple papers where we used a protocol similar to this and found hits. They
provide more specific methodological details:
Holden, P. M.; Kaur, H.; Gochin, M.; *Rizzo, R. C.* Footprint-based
identification of HIVgp41 inhibitors, *Bioorg. Med. Chem. Lett.*, *2012*,
Berger, W. T.; Ralph, B. P.; Kaczocha, M.; Sun, J.; Balius. T. E.; *Rizzo,
R. C.*; Haj-Dahmane, S.; Ojima, I.; Deutsch, D. G. Targeting Fatty Acid
Binding Protein (FABP) Anandamide Transporters - A Novel Strategy for
Development of Anti-Inflammatory and Anti-Nociceptive Drugs, *PLoS ONE*,
*2012*, *7*, e50968
Note that this is the product of several years of tweaking and refinement
and in-house scripts. The key is to find a pipeline that works for you
taking into account the resources you have available, your end goals, and
your target of interest. The footprint score, for example, may not be
useful if you don't have any structural information for known inhibitors.
It seems because of other activities you already have going on in your lab,
Amber score, as Scott said, is probably a bigger priority.
Feel free to run any other parameter sets by me if you are unsure of
anything. I would be happy to look them over and offer suggestions.
On Wed, Jun 18, 2014 at 4:56 PM, Scott Brozell <sbrozell at rci.rutgers.edu>
> On Wed, Jun 18, 2014 at 05:14:15PM +0000, Daniel Graham wrote:
> > I am currently using the dock input parameters shown on the Rizzo
> > I am currently participating in a docking study where we are using
> high throughput screening. I am screening millions of compounds using
> dock6's grid score function, however it yields too many hits to be useful.
> Is there a way I can be more selective?
> There are many approaches; see below for one.
> I'll let the tutorial writers make specific suggestions on their inputs.
> > ?We are also using Amber to perform detailed studies on molecules
> that show good potential, however some of my top dock scorers (-90 grid
> score) do not score as well as some of my average scorers (-40 grid score).
> It seems that after a molecule scores below -30 it might have potential.
> This presents an issue because Amber calculations take much longer and I
> don't want to have to go though thousands of compounds basically guessing
> about which ones may score well in Amber. Is there a way to generate grid
> scores that better reflect how the molecule will do in Amber?
> The general behavior that you state is typical.
> One step between Dock grid score and full MD in Amber would be
> Dock Amber score which is MM-GB/SA rescoring.
> The default input parameters for Dock Amber score could provide
> enough Amber minimization and MD to allow a superior ligand screen
> at a computational expense substantially more than grid score
> but substantially less than full MD.
> There are many caveats, but one important point is receptor preparation.
> If i infer correctly that you have already fully prepared your receptor
> for your full MD Amber calculations then you may have taken a big step
> towards eliminating one source of error in Dock Amber score.
> In addition, if you have specific knowledge about the receptor then a
> non-default input file using the NAB atom expression could be useful.
> For details see
> Dock-fans mailing list
> Dock-fans at docking.org
William Joseph Allen, Ph.D.
Dept. of Applied Mathematics and Statistics
Stony Brook University
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Dock-fans