[gmx-users] Using ATB to generate valid topologies

Post by ALEXANDER DHALIWAL
Dear GROMACS users,
I am pioneering the use of GROMACS in my lab to supplement some of our
experimental research. I have generated some topologies from the Automated
Topology Builder (ATB), but I am unsure whether or not to trust the files
provided by this service. I am only an undergrad, so I would not feel
comfortable asking my supervisor to invest thousands of dollars into a
program such as Gaussian09 to validate these topologies. Does this

One would not use Gaussian to validate ATB topologies. One would generally
*start* with a QM calculation to get charge distributions, then validate against
other data. The problem with the GROMOS force field is that there is little
connection between QM calculations and the final topologies. I have posted at
length on this topic just within the last few days; please check the archive.

Post by ALEXANDER DHALIWAL
community trust the outputs of ATB and, if not, could you provide me with
suggestions as to how I may validate my topologies given my restraints? Any
advice would be much appreciated.

It depends on how complex the molecules are. GROMOS parametrization is
empirical; existing building blocks are put together to yield a molecule, and
any unknown groups or strange linkages are explicitly parametrized using model
compounds. I don't recall all the guts of what ATB does (I haven't used the
GROMOS force field in a long time) but I believe it is doing this, coupled with
a quick QM calculation for charge refinement in the case that the constituent
groups are not easily looked up from an existing database). In general, the
topologies from ATB are significantly better than from, e.g. PRODRG, but the
skeptic/force field purist in me never trusts a black box. The problem, when it
comes to ligands/drugs is lack of target data. For small molecules, there are
thermodynamic properties that one can look at. For more complex things like
drugs, you might get lucky and have something like a logP tabulated somewhere,
but such data can often be difficult or impossible to obtain.

-Justin
--
==================================================

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

***@outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

==================================================

David van der Spoel

2016-07-12 12:40:37 UTC

One would not use Gaussian to validate ATB topologies. One would
generally *start* with a QM calculation to get charge distributions,
then validate against other data. The problem with the GROMOS force
field is that there is little connection between QM calculations and the
final topologies. I have posted at length on this topic just within the
last few days; please check the archive.

It depends on how complex the molecules are. GROMOS parametrization is
empirical; existing building blocks are put together to yield a
molecule, and any unknown groups or strange linkages are explicitly
parametrized using model compounds. I don't recall all the guts of what
ATB does (I haven't used the GROMOS force field in a long time) but I
believe it is doing this, coupled with a quick QM calculation for charge
refinement in the case that the constituent groups are not easily looked
up from an existing database). In general, the topologies from ATB are
significantly better than from, e.g. PRODRG, but the skeptic/force field
purist in me never trusts a black box. The problem, when it comes to
ligands/drugs is lack of target data. For small molecules, there are
thermodynamic properties that one can look at. For more complex things
like drugs, you might get lucky and have something like a logP tabulated
somewhere, but such data can often be difficult or impossible to obtain.
-Justin

XAvier Periole

2016-07-13 13:17:00 UTC

Just for the record ATB is a very good tool to build a start up topology. They also have a large set of topologies available on the site.

The charges are determined by QM and the user chooses the level of accuracy.
The bonded and LJ parameters are chosen based on the type of atom and it connectivity. Most atom types are already existing.

The sad part of ATB and other automated topology builders is that it is difficult to make a correct topology without putting the hands in.

Note that the CHARMM topology builder is even worst that ATB, but it gives you consistent topologies with existing ones.

In short: You need an experienced modeller to help you “validating” your topology.

X-

One would not use Gaussian to validate ATB topologies. One would
generally *start* with a QM calculation to get charge distributions,
then validate against other data. The problem with the GROMOS force
field is that there is little connection between QM calculations and the
final topologies. I have posted at length on this topic just within the
last few days; please check the archive.

It depends on how complex the molecules are. GROMOS parametrization is
empirical; existing building blocks are put together to yield a
molecule, and any unknown groups or strange linkages are explicitly
parametrized using model compounds. I don't recall all the guts of what
ATB does (I haven't used the GROMOS force field in a long time) but I
believe it is doing this, coupled with a quick QM calculation for charge
refinement in the case that the constituent groups are not easily looked
up from an existing database). In general, the topologies from ATB are
significantly better than from, e.g. PRODRG, but the skeptic/force field
purist in me never trusts a black box. The problem, when it comes to
ligands/drugs is lack of target data. For small molecules, there are
thermodynamic properties that one can look at. For more complex things
like drugs, you might get lucky and have something like a logP tabulated
somewhere, but such data can often be difficult or impossible to obtain.
-Justin

http://virtualchemistry.org provides validated topologies for about 150 compounds, that is force fields have been tested by running liquid simulations.
--
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone: +46184714205.
--
Gromacs Users mailing list
* Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
* For (un)subscribe requests visit

Justin Lemkul

2016-07-13 13:59:47 UTC

Post by XAvier Periole
Just for the record ATB is a very good tool to build a start up topology. They also have a large set of topologies available on the site.
The charges are determined by QM and the user chooses the level of accuracy.
The bonded and LJ parameters are chosen based on the type of atom and it connectivity. Most atom types are already existing.
The sad part of ATB and other automated topology builders is that it is difficult to make a correct topology without putting the hands in.
Note that the CHARMM topology builder is even worst that ATB, but it gives you consistent topologies with existing ones.

Which CHARMM topology builder? CGenFF or ParamChem? I'd be curious to know the
basis upon which this claim is made. CHARMM and GROMOS follow very different
parametrization protocols, so if a systematic analysis has been done, in the
context of the relative accuracy of the parent force fields, I would really like
to see the results. I think that would be a very valuable effort.

Note that at least CGenFF gives penalty scores for all parameters that it
provides the user, specifically telling which parameters will require additional
work.

Post by XAvier Periole
In short: You need an experienced modeller to help you “validating” your topology.

Agreed 100%. There is no black box method that is perfect or can be trusted
implicitly. Some are better than others, but force fields are as much an art as
they are a science. A perfect, hands-off method for parametrization of any
molecule is, at present, untenable.

-Justin

One would not use Gaussian to validate ATB topologies. One would
generally *start* with a QM calculation to get charge distributions,
then validate against other data. The problem with the GROMOS force
field is that there is little connection between QM calculations and the
final topologies. I have posted at length on this topic just within the
last few days; please check the archive.

It depends on how complex the molecules are. GROMOS parametrization is
empirical; existing building blocks are put together to yield a
molecule, and any unknown groups or strange linkages are explicitly
parametrized using model compounds. I don't recall all the guts of what
ATB does (I haven't used the GROMOS force field in a long time) but I
believe it is doing this, coupled with a quick QM calculation for charge
refinement in the case that the constituent groups are not easily looked
up from an existing database). In general, the topologies from ATB are
significantly better than from, e.g. PRODRG, but the skeptic/force field
purist in me never trusts a black box. The problem, when it comes to
ligands/drugs is lack of target data. For small molecules, there are
thermodynamic properties that one can look at. For more complex things
like drugs, you might get lucky and have something like a logP tabulated
somewhere, but such data can often be difficult or impossible to obtain.
-Justin

XAvier Periole

2016-07-13 15:02:28 UTC

Which CHARMM topology builder? CGenFF or ParamChem? I'd be curious to know the basis upon which this claim is made.

My experience in building CHARMM topologies … building from existing topologies is rather coarse!

CHARMM and GROMOS follow very different parametrization protocols, so if a systematic analysis has been done, in the context of the relative accuracy of the parent force fields, I would really like to see the results. I think that would be a very valuable effort.

Yes CGgenFF, it builds from building blocks and not clear to me how it adjusts the charges … but no QM!

Note that at least CGenFF gives penalty scores for all parameters that it provides the user, specifically telling which parameters will require additional work.

How does it determine them?
No penalty does not guaranty good parameters but small expected deviation from existing topology …

Post by XAvier Periole
In short: You need an experienced modeller to help you “validating” your topology.

Agreed 100%. There is no black box method that is perfect or can be trusted implicitly. Some are better than others, but force fields are as much an art as they are a science. A perfect, hands-off method for parametrization of any molecule is, at present, untenable.
-Justin

One would not use Gaussian to validate ATB topologies. One would
generally *start* with a QM calculation to get charge distributions,
then validate against other data. The problem with the GROMOS force
field is that there is little connection between QM calculations and the
final topologies. I have posted at length on this topic just within the
last few days; please check the archive.

It depends on how complex the molecules are. GROMOS parametrization is
empirical; existing building blocks are put together to yield a
molecule, and any unknown groups or strange linkages are explicitly
parametrized using model compounds. I don't recall all the guts of what
ATB does (I haven't used the GROMOS force field in a long time) but I
believe it is doing this, coupled with a quick QM calculation for charge
refinement in the case that the constituent groups are not easily looked
up from an existing database). In general, the topologies from ATB are
significantly better than from, e.g. PRODRG, but the skeptic/force field
purist in me never trusts a black box. The problem, when it comes to
ligands/drugs is lack of target data. For small molecules, there are
thermodynamic properties that one can look at. For more complex things
like drugs, you might get lucky and have something like a logP tabulated
somewhere, but such data can often be difficult or impossible to obtain.
-Justin

--
==================================================
Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow
Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201
http://mackerell.umaryland.edu/~jalemkul <http://mackerell.umaryland.edu/~jalemkul>
==================================================
--
Gromacs Users mailing list
* Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List <http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List> before posting!
* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists <http://www.gromacs.org/Support/Mailing_Lists>
* For (un)subscribe requests visit

Justin Lemkul

2016-07-13 15:39:02 UTC

Which CHARMM topology builder? CGenFF or ParamChem? I'd be curious to know the basis upon which this claim is made.

My experience in building CHARMM topologies … building from existing topologies is rather coarse!

Yes CGgenFF, it builds from building blocks and not clear to me how it adjusts the charges … but no QM!

Note that at least CGenFF gives penalty scores for all parameters that it provides the user, specifically telling which parameters will require additional work.

How does it determine them?
No penalty does not guaranty good parameters but small expected deviation from existing topology …

The answers to your questions above are published (particularly the second paper):

http://dx.doi.org/10.1021/ci300363c
http://dx.doi.org/10.1021/ci3003649

-Justin

Post by XAvier Periole
In short: You need an experienced modeller to help you “validating” your topology.

Agreed 100%. There is no black box method that is perfect or can be trusted implicitly. Some are better than others, but force fields are as much an art as they are a science. A perfect, hands-off method for parametrization of any molecule is, at present, untenable.
-Justin

One would not use Gaussian to validate ATB topologies. One would
generally *start* with a QM calculation to get charge distributions,
then validate against other data. The problem with the GROMOS force
field is that there is little connection between QM calculations and the
final topologies. I have posted at length on this topic just within the
last few days; please check the archive.

It depends on how complex the molecules are. GROMOS parametrization is
empirical; existing building blocks are put together to yield a
molecule, and any unknown groups or strange linkages are explicitly
parametrized using model compounds. I don't recall all the guts of what
ATB does (I haven't used the GROMOS force field in a long time) but I
believe it is doing this, coupled with a quick QM calculation for charge
refinement in the case that the constituent groups are not easily looked
up from an existing database). In general, the topologies from ATB are
significantly better than from, e.g. PRODRG, but the skeptic/force field
purist in me never trusts a black box. The problem, when it comes to
ligands/drugs is lack of target data. For small molecules, there are
thermodynamic properties that one can look at. For more complex things
like drugs, you might get lucky and have something like a logP tabulated
somewhere, but such data can often be difficult or impossible to obtain.
-Justin

XAvier Periole

2016-07-14 06:34:06 UTC

Tks for the references.

For the record, here are the references for ATB:
DOI: 10.1021/ct200196m <http://pubs.acs.org/doi/abs/10.1021/ct200196m>
DOI:10.1089/cmb.2012.0239 <http://dx.doi.org/10.1089/cmb.2012.0239>
DOI:10.1007/s10822-014-9713-7 <http://dx.doi.org/10.1007/s10822-014-9713-7>

X-

Which CHARMM topology builder? CGenFF or ParamChem? I'd be curious to know the basis upon which this claim is made.

My experience in building CHARMM topologies … building from existing topologies is rather coarse!

Yes CGgenFF, it builds from building blocks and not clear to me how it adjusts the charges … but no QM!

Note that at least CGenFF gives penalty scores for all parameters that it provides the user, specifically telling which parameters will require additional work.

How does it determine them?
No penalty does not guaranty good parameters but small expected deviation from existing topology …

http://dx.doi.org/10.1021/ci300363c <http://dx.doi.org/10.1021/ci300363c>
http://dx.doi.org/10.1021/ci3003649 <http://dx.doi.org/10.1021/ci3003649>
-Justin

Post by XAvier Periole
In short: You need an experienced modeller to help you “validating” your topology.

Agreed 100%. There is no black box method that is perfect or can be trusted implicitly. Some are better than others, but force fields are as much an art as they are a science. A perfect, hands-off method for parametrization of any molecule is, at present, untenable.
-Justin