ADL: Positve autodock score for a cristallographic

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

ADL: Positve autodock score for a cristallographic

Nicolas Sapay
Hello everyone,

I'm trying to calculate the binding score of a ligand-protein complex with
Autodock 4.2.1. The complex comes from a crystallographic structure to
which I have added hydrogen and partial charges. The autodock score is
calculated with the following dpf file:
  autodock_parameter_version 4.2  # used by autodock to validate
  parameter set
  outlev 0                        # diagnostic output level
  intelec                         # calculate internal electrostatics
  seed  pid time                   # seeds for random generator
  ligand_types HD C S OA N        # atoms types in ligand
  fld protein.maps.fld            # grid_data_file
  map protein.HD.map              # atom-specific affinity map
  map protein.C.map               # atom-specific affinity map
  map protein.S.map               # atom-specific affinity map
  map protein.OA.map              # atom-specific affinity map
  map protein.N.map               # atom-specific affinity map
  elecmap protein.e.map           # electrostatics map
  desolvmap protein.d.map         # desolvation map
  move ligand.pdbqt               # small molecule
  about 0.914 3.924 -16.686       # small molecule center
  epdb

The grid is center on the ligand and is sufficiently large to enclose the
ligand, although the ligand is pretty big with 42 torsions and highly
negatively charged (-12 e). The calculated score is +8.7 kcal/mol. I was
expecting something negative or at least close to 0, regarding the amount
of salt bridges between the ligand and the protein. The electrostatics is
negative (-13 kcal/mol) as  well as the final intermolecular energy (-9.1
kcal/mol), but this is compensated by large desolvation and torsional
energies. Actually, I'm not fully sure on how to interpret this results.
My guesses are:
  1. the scoring function/parameters are not suited for such a large ligand;
  2. the energy of the unbound state is not correctly evaluate (indeed, it
is 0 kcal/mol);
  3. the desolvation energy is not correctly evaluated because of the high
negative charge (something like an accumulation of small imprecision);
  4.  a combination of the three above.

Does someone have a suggestion? I should precise that my objective is not
to evaluate the binding energy of that complex. I just want to know if
other compounds can to better or worse than this particular ligand.

Thanks,
Nicolas



________________________________________________
--- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
Reply | Threaded
Open this post in threaded view
|

Re: ADL: Positve autodock score for a cristallographic

Mark Swingle
Nicolas,

The 4.2 default parameter set assumes that the unbound energy = docked ligand internal energy. It seems like epdb is not handling that correctly? What do you get if you subtract the internal energy? Also, the torsional term is very simple: ntor X constant. It's a very rough approximation that may not be appropriate for an unusual ligand like this. You might also want to try to relax the crystallographic conformation WRT autodock's scoring function with a local search.

Regards,

Mark

----- Original Message -----
From: Nicolas SAPAY <[hidden email]>
Date: Monday, August 10, 2009 5:03 am
Subject: ADL: Positve autodock score for a cristallographic
To: [hidden email]


> Hello everyone,
>
> I'm trying to calculate the binding score of a ligand-protein complex
> with
> Autodock 4.2.1. The complex comes from a crystallographic structure to
> which I have added hydrogen and partial charges. The autodock score is
> calculated with the following dpf file:
>   autodock_parameter_version 4.2  # used by autodock to validate
>   parameter set
>   outlev 0                        # diagnostic output level
>   intelec                         # calculate internal electrostatics
>   seed  pid time                   # seeds for random generator
>   ligand_types HD C S OA N        # atoms types in ligand
>   fld protein.maps.fld            # grid_data_file
>   map protein.HD.map              # atom-specific affinity map
>   map protein.C.map               # atom-specific affinity map
>   map protein.S.map               # atom-specific affinity map
>   map protein.OA.map              # atom-specific affinity map
>   map protein.N.map               # atom-specific affinity map
>   elecmap protein.e.map           # electrostatics map
>   desolvmap protein.d.map         # desolvation map
>   move ligand.pdbqt               # small molecule
>   about 0.914 3.924 -16.686       # small molecule center
>   epdb
>
> The grid is center on the ligand and is sufficiently large to enclose
> the
> ligand, although the ligand is pretty big with 42 torsions and highly
> negatively charged (-12 e). The calculated score is +8.7 kcal/mol. I was
> expecting something negative or at least close to 0, regarding the amount
> of salt bridges between the ligand and the protein. The electrostatics
> is
> negative (-13 kcal/mol) as  well as the final intermolecular energy (-9.1
> kcal/mol), but this is compensated by large desolvation and torsional
> energies. Actually, I'm not fully sure on how to interpret this results.
> My guesses are:
>   1. the scoring function/parameters are not suited for such a large ligand;
>   2. the energy of the unbound state is not correctly evaluate
> (indeed, it
> is 0 kcal/mol);
>   3. the desolvation energy is not correctly evaluated because of the
> high
> negative charge (something like an accumulation of small imprecision);
>   4.  a combination of the three above.
>
> Does someone have a suggestion? I should precise that my objective is
> not
> to evaluate the binding energy of that complex. I just want to know if
>
> other compounds can to better or worse than this particular ligand.
>
> Thanks,
> Nicolas
>
>
>
> ________________________________________________
> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
________________________________________________
--- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
Reply | Threaded
Open this post in threaded view
|

Re: ADL: Positve autodock score for a cristallographic

Nicolas Sapay
Thanks for you answer,

> Nicolas,
>
> The 4.2 default parameter set assumes that the unbound energy = docked
> ligand internal energy. It seems like epdb is not handling that correctly?

Actually, Autodock says:
Total Intramolecular Interaction Energy   = +5.252 kcal/mol
Unbound model to be used is 'same as bound' [AutoDock 4.1 default].
...
Total Intermolecular Interaction Energy   = -3.833 kcal/mol
Total Intermolecular vdW + Hbond Energy   = +3.950 kcal/mol
Total Intermolecular Electrostatic Energy = -13.034 kcal/mol

epdb: USER    Estimated Free Energy of Binding    =   +8.70 kcal/mol
[=(1)+(2)+(3)-(4)]
epdb: USER
epdb: USER    (1) Final Intermolecular Energy     =   -9.08 kcal/mol
epdb: USER        vdW + Hbond + desolv Energy     =   +3.95 kcal/mol
epdb: USER        Electrostatic Energy            =  -13.03 kcal/mol
epdb: USER    (2) Final Total Internal Energy     =   +5.25 kcal/mol
epdb: USER    (3) Torsional Free Energy           =  +12.53 kcal/mol
epdb: USER    (4) Unbound System's Energy         =   +0.00 kcal/mol
epdb: USER
epdb: USER

So, if I understand well, Autodock does not really evaluate the energy of
the unbound system. Or those it means that as total energy of the bound
and the unbound systems are considered the same, the final energy is not
modified? I'm a bit confused... Intuitively, I would use the Autodock
score and the sum between (1) and (2) as references for the other
compounds I want to test, assuming their conformation and number of
torsions are close to the crystallographic structure.

> What do you get if you subtract the internal energy? Also, the torsional
> term is very simple: ntor X constant. It's a very rough approximation that
> may not be appropriate for an unusual ligand like this. You might also

Indeed, the torsional energy might be off by few kcal.

> want to try to relax the crystallographic conformation WRT autodock's
> scoring function with a local search.

That's a good idea. The complex is a bit "stressed" in one area close to
the core of the binding site. I have observed that when minimizing the
complex to prepare a MD simulation.

Regards,
Nicolas

>
> Regards,
>
> Mark
>
> ----- Original Message -----
> From: Nicolas SAPAY <[hidden email]>
> Date: Monday, August 10, 2009 5:03 am
> Subject: ADL: Positve autodock score for a cristallographic
> To: [hidden email]
>
>
>> Hello everyone,
>>
>> I'm trying to calculate the binding score of a ligand-protein complex
>> with
>> Autodock 4.2.1. The complex comes from a crystallographic structure to
>> which I have added hydrogen and partial charges. The autodock score is
>> calculated with the following dpf file:
>>   autodock_parameter_version 4.2  # used by autodock to validate
>>   parameter set
>>   outlev 0                        # diagnostic output level
>>   intelec                         # calculate internal electrostatics
>>   seed  pid time                   # seeds for random generator
>>   ligand_types HD C S OA N        # atoms types in ligand
>>   fld protein.maps.fld            # grid_data_file
>>   map protein.HD.map              # atom-specific affinity map
>>   map protein.C.map               # atom-specific affinity map
>>   map protein.S.map               # atom-specific affinity map
>>   map protein.OA.map              # atom-specific affinity map
>>   map protein.N.map               # atom-specific affinity map
>>   elecmap protein.e.map           # electrostatics map
>>   desolvmap protein.d.map         # desolvation map
>>   move ligand.pdbqt               # small molecule
>>   about 0.914 3.924 -16.686       # small molecule center
>>   epdb
>>
>> The grid is center on the ligand and is sufficiently large to enclose
>> the
>> ligand, although the ligand is pretty big with 42 torsions and highly
>> negatively charged (-12 e). The calculated score is +8.7 kcal/mol. I was
>> expecting something negative or at least close to 0, regarding the
>> amount
>> of salt bridges between the ligand and the protein. The electrostatics
>> is
>> negative (-13 kcal/mol) as  well as the final intermolecular energy
>> (-9.1
>> kcal/mol), but this is compensated by large desolvation and torsional
>> energies. Actually, I'm not fully sure on how to interpret this results.
>> My guesses are:
>>   1. the scoring function/parameters are not suited for such a large
>> ligand;
>>   2. the energy of the unbound state is not correctly evaluate
>> (indeed, it
>> is 0 kcal/mol);
>>   3. the desolvation energy is not correctly evaluated because of the
>> high
>> negative charge (something like an accumulation of small imprecision);
>>   4.  a combination of the three above.
>>
>> Does someone have a suggestion? I should precise that my objective is
>> not
>> to evaluate the binding energy of that complex. I just want to know if
>>
>> other compounds can to better or worse than this particular ligand.
>>
>> Thanks,
>> Nicolas
>>
>>
>>
>> ________________________________________________
>> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
> ________________________________________________
> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>


________________________________________________
--- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
Reply | Threaded
Open this post in threaded view
|

Re: ADL: Positve autodock score for a cristallographic

Stefano Forli
Hi Nicolas,

looking at the details of the energy calculation, a culprit for your positive energy seems
to be the entropy penalty (Torsions Free Energy) associated with the unusual number of
torsions. The AutoDock scoring function provides an additive term for each torsion that
was tuned to work with relatively small ligands.

What you could do is to reduce value associated with the torsion penalty in the AutoDock
parameters files (AD*.dat):

FE_coeff_tors   0.2xxxx

and either recompile AutoDock again or provide the custom file in your calculations using
the "parameter_file FILENAME" keyword in the DPF. Although in order to use it in your
calculaitons you should make some validations of the new value, possibly with a known
binder with similar number of torsions.
By the way, you can consider this as a "systematic error" that shouldn't bias the results
in any direction.

On the other hand, the high number of charges could affect the accuracy of your search by
encouraging electrostatic interactions over the non-polar ones.
There are examples in which the partial charges were successfully reduced for heavily
charged system (Moitessier N, Westhof E, Hanessian S., J Med Chem. 2006,9;49(3):1023-33)

Hope this helps,

Stefano

Nicolas SAPAY wrote:

> Thanks for you answer,
>
>> Nicolas,
>>
>> The 4.2 default parameter set assumes that the unbound energy = docked
>> ligand internal energy. It seems like epdb is not handling that correctly?
>
> Actually, Autodock says:
> Total Intramolecular Interaction Energy   = +5.252 kcal/mol
> Unbound model to be used is 'same as bound' [AutoDock 4.1 default].
> ...
> Total Intermolecular Interaction Energy   = -3.833 kcal/mol
> Total Intermolecular vdW + Hbond Energy   = +3.950 kcal/mol
> Total Intermolecular Electrostatic Energy = -13.034 kcal/mol
>
> epdb: USER    Estimated Free Energy of Binding    =   +8.70 kcal/mol
> [=(1)+(2)+(3)-(4)]
> epdb: USER
> epdb: USER    (1) Final Intermolecular Energy     =   -9.08 kcal/mol
> epdb: USER        vdW + Hbond + desolv Energy     =   +3.95 kcal/mol
> epdb: USER        Electrostatic Energy            =  -13.03 kcal/mol
> epdb: USER    (2) Final Total Internal Energy     =   +5.25 kcal/mol
> epdb: USER    (3) Torsional Free Energy           =  +12.53 kcal/mol
> epdb: USER    (4) Unbound System's Energy         =   +0.00 kcal/mol
> epdb: USER
> epdb: USER
>
> So, if I understand well, Autodock does not really evaluate the energy of
> the unbound system. Or those it means that as total energy of the bound
> and the unbound systems are considered the same, the final energy is not
> modified? I'm a bit confused... Intuitively, I would use the Autodock
> score and the sum between (1) and (2) as references for the other
> compounds I want to test, assuming their conformation and number of
> torsions are close to the crystallographic structure.
>
>> What do you get if you subtract the internal energy? Also, the torsional
>> term is very simple: ntor X constant. It's a very rough approximation that
>> may not be appropriate for an unusual ligand like this. You might also
>
> Indeed, the torsional energy might be off by few kcal.
>
>> want to try to relax the crystallographic conformation WRT autodock's
>> scoring function with a local search.
>
> That's a good idea. The complex is a bit "stressed" in one area close to
> the core of the binding site. I have observed that when minimizing the
> complex to prepare a MD simulation.
>
> Regards,
> Nicolas
>
>> Regards,
>>
>> Mark
>>
>> ----- Original Message -----
>> From: Nicolas SAPAY <[hidden email]>
>> Date: Monday, August 10, 2009 5:03 am
>> Subject: ADL: Positve autodock score for a cristallographic
>> To: [hidden email]
>>
>>
>>> Hello everyone,
>>>
>>> I'm trying to calculate the binding score of a ligand-protein complex
>>> with
>>> Autodock 4.2.1. The complex comes from a crystallographic structure to
>>> which I have added hydrogen and partial charges. The autodock score is
>>> calculated with the following dpf file:
>>>   autodock_parameter_version 4.2  # used by autodock to validate
>>>   parameter set
>>>   outlev 0                        # diagnostic output level
>>>   intelec                         # calculate internal electrostatics
>>>   seed  pid time                   # seeds for random generator
>>>   ligand_types HD C S OA N        # atoms types in ligand
>>>   fld protein.maps.fld            # grid_data_file
>>>   map protein.HD.map              # atom-specific affinity map
>>>   map protein.C.map               # atom-specific affinity map
>>>   map protein.S.map               # atom-specific affinity map
>>>   map protein.OA.map              # atom-specific affinity map
>>>   map protein.N.map               # atom-specific affinity map
>>>   elecmap protein.e.map           # electrostatics map
>>>   desolvmap protein.d.map         # desolvation map
>>>   move ligand.pdbqt               # small molecule
>>>   about 0.914 3.924 -16.686       # small molecule center
>>>   epdb
>>>
>>> The grid is center on the ligand and is sufficiently large to enclose
>>> the
>>> ligand, although the ligand is pretty big with 42 torsions and highly
>>> negatively charged (-12 e). The calculated score is +8.7 kcal/mol. I was
>>> expecting something negative or at least close to 0, regarding the
>>> amount
>>> of salt bridges between the ligand and the protein. The electrostatics
>>> is
>>> negative (-13 kcal/mol) as  well as the final intermolecular energy
>>> (-9.1
>>> kcal/mol), but this is compensated by large desolvation and torsional
>>> energies. Actually, I'm not fully sure on how to interpret this results.
>>> My guesses are:
>>>   1. the scoring function/parameters are not suited for such a large
>>> ligand;
>>>   2. the energy of the unbound state is not correctly evaluate
>>> (indeed, it
>>> is 0 kcal/mol);
>>>   3. the desolvation energy is not correctly evaluated because of the
>>> high
>>> negative charge (something like an accumulation of small imprecision);
>>>   4.  a combination of the three above.
>>>
>>> Does someone have a suggestion? I should precise that my objective is
>>> not
>>> to evaluate the binding energy of that complex. I just want to know if
>>>
>>> other compounds can to better or worse than this particular ligand.
>>>
>>> Thanks,
>>> Nicolas
>>>
>>>
>>>
>>> ________________________________________________
>>> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>> ________________________________________________
>> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>>
>
>
> ________________________________________________
> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---


--
  Stefano Forli, PhD

  Research Associate
  Olson Molecular Graphics Laboratory
  Dept. Molecular Biology,  MB-5
  The Scripps Research Institute
  10550  North Torrey Pines Road
  La Jolla,  CA 92037-1000,  USA.

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