[gmx-users] NOE restraints and undefined hydrogen atoms, virtual interaction sites

[Mark Abraham]

chris.neale at utoronto.ca wrote:
>> There's no a priori reason why a mixture of a united-atom detergent 
>> sodium dodecyl sulfate and an all-atom peptide *couldn't* work. 
>> Neither is there any reason to suppose they *would* work without 
>> evidence that the SDS parameters were developed with this purpose in 
>> mind and suitably validated. For that, you'll have to read the paper 
>> referred to on the page you footnoted [1], I guess. Your other 
>> alternative is looking for an all-atom SDS topology that has been 
>> validated, but I've no idea whether one exists.
> 
> Do you have any suggestions how one would validate the combination of 
> two such force-fields? While generally one could always develop new 
> parameters using the same methodology as the existing forcefield, it is 
> not obvious how one would do this when combining all-atom and 
> united-atom representations.

No, I don't have specific ideas - I've never considered the issue. All I 
think that one could do is to make the decisions for that particular 
combination in a way that might be thought sensible, and to test the 
results against experimental data.

> I see how the basic combination rules, etc. could be validated by 
> simulating for example an aqueous protein restrained to one half of the 
> simulation box and an SDS micelle restrained to the other half of the 
> box in order to ensure no vdw contacts and then compare equilibrium 
> properties from this simulation to those of an aqueous protein and an 
> SDS micelle in separate simulations.

Hmmm, if you've arranged no vdw contacts between the two pieces of force 
field, then you're really not testing anything much with the above. 
However failure with the above should indicate you should go and work on 
something easier, like modelling brain function. :-)

> Further, in some situations one could look at the parameters to see the 
> there are no striking differences between similar atom types. This would 
> definitely be useful for example when trying to combine an all-atom 
> protein forcefield with a united-atom forcefield. However, SDS doesn't 
> really have any groups that are similar to and protein groups. Further, 
> it is not readily apparent to me how one would even extend a protein 
> forcefield using the same methodology to SDS and be sure based simply on 
> methodology of generation that they were valid in combination since 
> there is so little overlap in actual group types.

True.

Mark