ADL: term"randomized structure"

Stefano Forli forli at scripps.edu
Sat Mar 26 12:32:41 PDT 2016


Richard,
your reply sparks interesting discussion topics.

>I don't agree with the "within 2.0 A range you can assume that most of the known essential interactions with the target have been
>established" comment.  That would mean that a molecule such as geldanamycin could have completely flipped its orientation.
I'm not sure this would be the case, actually.
By flipping a 19-membered ring macrocycle such as geldanamycin, the RMSD would be very likely in the two-digit realm.
On the other hand, 2.0A RMSD could likely mean that the core of a molecule is right on spot, but a floppy chain with no specific interactions is deviating from the experimental coordinates.

>>"Several papers showed redocking ligand structures generated from SMILES can reduce docking
>>accuracy by >20%."
>I suspect a person that did that would be an organic chemist.
>In my experience, a program like ChemAxon does NOT do a "complete optimization" like say, Gaussian would.
I'm not sure that being an organic chemist is a such bad thing :)
One of the papers I've linked in my previous reply is actually reporting these figures. Not sure about the organic chemistry skills of the authors, but they're members of the CCDC (and half of them are the authors of the GOLD docking software).
Programs such as ChemAxon-Marvin, Corina or Omega are wonderful tools that generate 3D coordinates in milliseconds, compared to minutes or hours for any QM software. They don't do a real optimization, but using look-up tables they generate extremely accurate structures (again: bond lengths and angles).
Although, the point is quite the opposite. Docking success rate could be lower because input structures could be much closer to the ideal values than the experimental coordinates in the crystal structure (which can be distorted). Paradoxically, this means that even if Gaussian would generate better structures, it wouldn't improve docking results (and possibly, make them worst). 
This a minor (but not negligible) component of the induced fit upon binding.

>>"In fact, pretty much every docking software (including AutoDock) do not alter bond angles
>>and lengths during the calculation"
>If so, how do programs such as SurflexDock do binding site optimizations?  Surely they must read in a force field.
I totally agree with you: pretty much every docking software, not all of them.

>Sorry, I'm a computational chemist, and I take this kind of stuff VERY seriously because there are people with far less training than I have that are more gainfully employed than I >am.  There are many people working in pharma that were "trained" by the companies they work for to be "modelers" because they are cheaper.
>
>Richard


--

 Stefano Forli, PhD

 Assistant Professor of Integrative
 Structural and Computational Biology,
 Molecular Graphics Laboratory

 Dept. of Integrative Structural
  and Computational Biology, MB-112F
 The Scripps Research Institute
 10550  North Torrey Pines Road
 La Jolla,  CA 92037-1000,  USA.

    tel: +1 (858)784-2055
    fax: +1 (858)784-2860
    email: forli at scripps.edu
    http://www.scripps.edu/~forli/
________________________________________
From: autodock-bounces at scripps.edu [autodock-bounces at scripps.edu] On Behalf Of Richard  Wood [Richard.Wood at purduecal.edu]
Sent: Saturday, March 26, 2016 9:51 AM
To: autodock at scripps.edu
Subject: Re: ADL: term"randomized structure"

I don't agree with the "within 2.0 A range you can assume that most of the known essential interactions with the target have been
established" comment.  That would mean that a molecule such as geldanamycin could have completely flipped its orientation.

"Several papers showed redocking ligand structures generated from SMILES can reduce docking
accuracy by >20%."

I suspect a person that did that would be an organic chemist.

In my experience, a program like ChemAxon does NOT do a "complete optimization" like say, Gaussian would.

"In fact, pretty much every docking software (including AutoDock) do not alter bond angles
and lengths during the calculation"

If so, how do programs such as SurflexDock do binding site optimizations?  Surely they must read in a force field.

Sorry, I'm a computational chemist, and I take this kind of stuff VERY seriously because there are people with far less training than I have that are more gainfully employed than I am.  There are many people working in pharma that were "trained" by the companies they work for to be "modelers" because they are cheaper.

Richard
________________________________________
From: autodock-bounces at scripps.edu [autodock-bounces at scripps.edu] on behalf of Stefano Forli [forli at scripps.edu]
Sent: Friday, March 25, 2016 1:42 PM
To: autodock at scripps.edu
Subject: Re: ADL: term"randomized structure"

Hi Richard,
just to clarify, if you create your molecule from scratch (e.g., you sketch it from a 2D
programs like ChemDraw) you want to carefully prepare the 3D coordinates, so you *must* do
a complete minimization.

In fact, pretty much every docking software (including AutoDock) do not alter bond angles
and lengths during the calculation, so if your structure is distorted or not optimal, it
will lead to poor/wrong results.
Several papers showed redocking ligand structures generated from SMILES can reduce docking
accuracy by >20%.

About the RMSD cutoff, 2.0 A is widely accepted as the standard for evaluating docking
performance, even though there's strong agreement on RMSD itself being a poor metric for
docking accuracy.
Alternatives get constantly proposed, but despite its limitations, within 2.0 A range you
can assume that most of the known essential interactions with the target have been
established.

Best,

S

On 03/25/2016 06:03 AM, Richard  Wood wrote:
> Generally, unless one uses a 2-D drawing program, most sketching programs will do some kind of "quick and dirty" optimization.  So, when I said, "non-optimized", that is what I was referring to.  Of course, one can also do a full blown optimization and dock that as well.
>
> It is probably best to remove the ligand from a ligand-protein crystal structure and re-dock the ligand to see if one can place it in its correct position.  One can then tweak docking parameters to see if they can reproduce the crystal ligand pose.  You can then calculate the RMSD between your docked ligand and that of the crystal structure ligand.  If one is careful, you can get an RMSD of less than 0.5.  (As an aside, I was at an ACS talk a few years back, and there was a speaker from some pharma company who was re-docking ligands and comparing them to the crystal.  He was proud of the fact that his RMSDs were less than 2.  I told him after his talk that he could do much better than that (he I was a lowly post-doc and he was a "professional modeler"-I still remain far below "professional modeler").)
>
> It's questionable how many people actually do that, however.
>
> Richard
> ________________________________________
> From: autodock-bounces at scripps.edu [autodock-bounces at scripps.edu] on behalf of Stefano Forli [forli at scripps.edu]
> Sent: Thursday, March 24, 2016 8:17 PM
> To: autodock at scripps.edu
> Subject: Re: ADL: term"randomized structure"
>
> Hi,
> a general practice when testing docking programs is to prevent any kind of bias in the
> input structures that can simplify the search and artificially increase re-docking success
> rate.
>
> For this reason, the input structures should be randomized to be different from the
> solution, but still optimized, i.e., all bond angles and lengths must be correct.
>
> The ideal approach would be to generate the 3D coordinates of the ligand starting from
> scratch (e.g., SMILES format) using an external software and dock that.
>
> Although, since during docking bond angles and lengths are not altered, it is sufficient
> to randomize the conformation of the ligand from its experimental coordinates in a
> crystallographic complex, by altering its position, orientation and all the angles in
> rotatable bonds.
>
>
> S
>
>
>
> On 03/24/2016 06:09 PM, Richard  Wood wrote:
>> Non-optimized?
>>
>> Richard
>>
>> ________________________________________
>> From: autodock-bounces at scripps.edu [autodock-bounces at scripps.edu] on behalf of maria timoshik [m.timoshik at yandex.ru]
>> Sent: Thursday, March 24, 2016 2:30 PM
>> To: autodock at scripps.edu
>> Subject: ADL: term"randomized structure"
>>
>> Hello:
>> I've heard that in a very beginning of video tutorial to Vina Dr. Oleg Todd said:
>> "I'll use a randomized structure of a drug molecule".
>> What the term RANDOMIZED STRUCTURE actually mean in this context?
>> Perhaps professional physico-chemists can explain?
>>
>> with regards, Maria
>> ________________________________________________
>> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>>
>> ________________________________________________
>> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>>
>
> --
>
>    Stefano Forli, PhD
>
>    Assistant Professor of Integrative
>    Structural and Computational Biology,
>    Molecular Graphics Laboratory
>
>    Dept. of Integrative Structural
>     and Computational Biology, MB-112A
>    The Scripps Research Institute
>    10550  North Torrey Pines Road
>    La Jolla,  CA 92037-1000,  USA.
>
>       tel: +1 (858)784-2055
>       fax: +1 (858)784-2860
>       email: forli at scripps.edu
>       http://www.scripps.edu/~forli/
> ________________________________________________
> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>
> ________________________________________________
> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>

--

  Stefano Forli, PhD

  Assistant Professor of Integrative
  Structural and Computational Biology,
  Molecular Graphics Laboratory

  Dept. of Integrative Structural
   and Computational Biology, MB-112A
  The Scripps Research Institute
  10550  North Torrey Pines Road
  La Jolla,  CA 92037-1000,  USA.

     tel: +1 (858)784-2055
     fax: +1 (858)784-2860
     email: forli at scripps.edu
     http://www.scripps.edu/~forli/
________________________________________________
--- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---

________________________________________________
--- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---


More information about the autodock mailing list