Re: AW: Dihedrals multiplicity!

From: Kenno Vanommeslaeghe (kvanomme_at_rx.umaryland.edu)
Date: Mon Feb 17 2014 - 12:44:19 CST

Simulations started from protein X-ray structures are the most common
source of publications with bad protonation states, because the hydrogens
are not resolved and some programs just assign them the same for each
instance of the same amino acid, which is dead wrong. However, this trap
can easily be avoided by using one of the many preparation tools that make
an educated guess for the amino acid protonation states (though catalytic
sites usually need to be considered manually nevertheless). Organic
ligands are more problematic in that respect, because if they have a
protonatable/tautomeric group, it's often more difficult to predict their
protonation state. To make matters worse, the results of getting it wrong
are often catastrophic when it comes to studying small molecule binding.
That said, there's not much you, as a toolmaker, can do. As you say,
tautomers are tough; it would be a project in its own right to get them
right programmatically for a descent percentage of cases. Unless you can
find some library that does so, I think the sanest course of action is to
issue big fat warnings to your users that they need to carefully consider
protonation and tautomeric states. That's what we're doing with paramchem,
though the many users that ignore our warnings are a great source of
frustration.

On 02/17/2014 12:43 PM, hannes.loeffler_at_stfc.ac.uk wrote:
> Hi Kenno,
>
> when you say tautomers and protonation state do you mean proteins or also organic ligands? I am writing an automatic setup tool (currently primarily for alchemical free energy calculations) and so I am looking for ways to automate this, as far as it is possible, too. Tautomers are tough!
>
> Cheers,
> Hannes.
>
> ________________________________________
> From: Kenno Vanommeslaeghe [kvanomme_at_rx.umaryland.edu]
> Sent: 17 February 2014 16:56
> To: namd-l_at_ks.uiuc.edu
> Subject: Re: AW: namd-l: Dihedrals multiplicity!
>
> Hi Hannes,
>
> I thought so, but I wanted to use the opportunity to issue a warning to
> the community at large. Inappropriate use of force fields is one of the
> most common reasons for me to review papers negatively (though still not
> as common as failure to intelligently assign tautomeric and protonation
> states during system preparation).
>
> Cheers,
>
> Kenno.
>
>
> On 02/17/2014 05:15 AM, hannes.loeffler_at_stfc.ac.uk wrote:
>> Hi again,
>>
>> many thanks for your insights.
>>
>> I should have probably said that I have seen this in a publication and I do share your concerns regarding mixing force fields.
>>
>> Cheers,
>> Hannes.
>>
>> ________________________________________
>> From: Kenno Vanommeslaeghe [kvanomme_at_rx.umaryland.edu]
>> Sent: 16 February 2014 18:45
>> To: namd-l_at_ks.uiuc.edu
>> Subject: Re: AW: namd-l: Dihedrals multiplicity!
>>
>> On 02/16/2014 03:34 AM, hannes.loeffler_at_stfc.ac.uk wrote:
>>> Just out of interest: what would you think of a study that embeds a AMBER protein in a CHARMM lipid membrane bilayer?
>>
>> Depends. If the study is only in a planning/exploratory stage, I would say
>> "whichever advantage one hopes to get out of of mixing force fields, it's
>> not worth the uncertainty it casts over the results." If only a few weeks
>> worth of time had been invested in the mixed force field study, I would
>> say "one should consider redoing the calculations with consistent force
>> fields". If restarting is not an option because too much time has been
>> invested in the simulations and analysis, well, there are a few easy
>> things that can be done and said to make the results more
>> credible/palatable to knowledgeable reviewers. Though it also depends on
>> the purpose of the study; if it explicitly focuses on, for instance,
>> membrane properties at the interface with the protein, or order/disorder
>> of the protein induced by the presence of the membrane, then little can be
>> done to justify mixing force fields.(*) Finally, if you're talking about a
>> study you came across in the literature, I'd say "interpret with care". On
>> typical simulation time scales, the inaccuracies induced by the imbalance
>> of the nonbonded interactions are likely to be not catastrophic.
>> Properties that are very sensitive to this imbalance, as in the above
>> examples, are highly suspect. Conversely, phenomena that happen on the
>> inside of a rigid transmembrane protein may not be affected by the
>> membrane model much.
>>
>>
>>
>> (*) I should mention for the sake of completeness that one could
>> hypothetically just submit a paper and speculate none of the reviewers
>> will notice the error. Judging by the number of published MD studies with
>> serious methodological flaws I come across on a monthly basis, there's a
>> fair chance of getting get away with it. Knowingly doing so would of
>> course not be a very ethical course of action, and would contribute to the
>> bad reputation of our field, thus polluting the water for all of us.
>>
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:08 CST