[gmx-users] Re: [gmx-developers] Hardcoded (bad) parameters in dummy code...

[Erik Lindahl]

Hi,

> 
> But, if I remember correctly, for '-dummy hydrogens', *all* parameters
> are calculated from forcefield parameters. Only for the 'aromatics'
> all parameters and atom names are hard-coded; if you try to work
> that out starting from the bare aromatic sidechain topology and
> assorted parameters, thing *really* get complicated... ;-(

Sorry if I sounded too hard - I was just quite tired after fixing a lot 
of undocumented and non-standard stuff in the OPLS parameters and 
articles, so I didn't get too happy when I found similar stuff in our 
own code :-)

> 
> The atom type is only checked for "NL", which is the type for
> an umbrella-shaped -NH2 group, but nothing in the dummy code should 
> depend on the input coordinates, only on some of the forcefield 
> parameters.
> 

OK - The initial position of the MCH3/MNH3 dummy masses are calculated 
from the input coordinates, but since there are constrainttypes defined 
for them it won't matter. Good - this means that it doesn't depend on 
any coordinates, apart from the dummy aromatics parameters....

>> I'm not sure what to do with the aromatic stuff - unless it can be
>>solved trivially I will disable dummy aromatics for now
> 
> Must you be so drastic? It works the way it is, it is consistent at 
> least with ffgmx and ffgmx2, and the differences with ffG43a are
> realy very minimal (if at all, I haven't re-checked). We should
> however ofcourse issue a clear warning about which parameters are
> used. Wouldn't that be a good alternative to tossing?

I won't disable anything in the stable branch, but it must be fixed 
since the current setup is dangerous. One way could be to introduce a 
trivial database with entries for each NH3/CH3 group and the next heave 
atomtype and say which dummy mass should be used, or if it's planar.

I'm not sure what to do about the aromatics; TRP and HIS are quite a bit
more complicated in most other forcefields since not all bonds are equal
in the 5-rings. One alternative might be to replace the current 
hardcoded parameters with experimental values of indole & isoxazole ring 
geometry. That should be an improvement for G96 and Gromacs forcefields 
too...

In the development tree I think we should replace it completely 
(otherwise we don't have any incitament to fix it) and start writing a 
dummy database.

An alternative might be to move all the automatic dummy stuff to 
grompp/mdrun to hide it from the topology...

Cheers,

Erik