On 2/23/24 2:49 AM, Steinbrecher, Thomas via AMBER wrote:
> [External Email]
>
> Hi all,
>
> thanks for posting these comments Skanda, I was curious if you couldn't
> also take a "glass half-full" perspective on this:
>
> If you care about a residue that can have two protonation states, a single
> structure approach like pKa-ANI run on both states will give you an upper
> and lower bound of the pKa (or pKa shift relative to solvent), because the
> protonated simulation generates an ensemble that favors the protonated
> state, biasing the pKa to a more basic value, while the deprotonated
> simulation generates an ensemble that favors the deprotonated one, with the
> 'correct' pKa shift somewhere in between both results (and, in the limit of
> infinite simulation time, both results converging to the same value because
> the ensembles are both fully sampled in both simulations). This is not
> enough to decide where in the middle of the two results the correct pKa is,
> because we do not know the free energy difference between the two ensembles
> (and have no easy way to calculate that without coupling protonation state
> and conformation changes in a cpHMD run) ), but the pKa-ANI calculation
> gives you an indication how large the uncertainty in the pKa shift is, so
> you could consider it a (cheaper&faster?) way to figure out if a cpHMD
> simulation will be necessary.
>
> Kind Regards,
>
> Thomas
Hi Thomas.
Let me try to explain with an extreme example why what you propose would
not work.
Running MD with a residue protonated all the time, would be equivalent
to an experiment done at very low pH, at least 2-3 pH units below the
residue's pKa, which in principle is unknown. If you run it
unprotonated, you are running it 2-3 units ABOVE the unknown pKa.
Now, let's take a system that at very low and very high pH have very
different conformations. Which one would you use ? The limits are
actually opposite of what simple intuition would tell you, and the
bounds for pKa can easily span 6-7 pKa units in those cases.
Take a look at Tanford, C. J. Am. Chem. Soc. 1961, 83 (7), 1628 for a
very simple model of coupled conformation/pH equilibria where you can
see that if a system undergoes any type of conformational change
associated with pH, then you cannot study just protonation or
conformation, you must study them coupled.
Adrian
>
>
>
> On Fri, Feb 23, 2024 at 2:16 AM Skanda Sastry via AMBER <amber.ambermd.org>
> wrote:
>
>> Hello all,
>>
>> I just wanted to circle back and amend my comments about pKa-ANI - while it
>> was shown to outperform propka and give decent RMSEs for most residues in
>> their paper, talking with Professor Roitberg (who advised on pKa-ANI and
>> was an author of the ANI methods) gave me more clarity on an important
>> detail: For complicated cases with shifted pKas, a single-structure pKa
>> method like pKa-ANI is only able to sample one of the conformational
>> states. However, this leads to a paradox because you need the pKa to get
>> the correct protonation state, but you can't trust the pKa unless you run
>> the MD at the correct protonation state. I just wanted to make sure that
>> this detail was mentioned in the mailing list and archives because it gave
>> me a better understanding of cpH-MD (hopefully it can do the same for
>> others).
>>
>> To quote Prof. Roitberg directly:
>>
>> If you look closely at the paper, you will see that most of the
>> experimental pKas are very close to the values in solution, which makes
>> sense. The interesting prediction are for those that have larger
>> deviations, such as ASP 26 in Thioredoxin, which is also mentioned in the
>> paper.
>>
>> As written, the protocol was this: do regular pKa-ANI, and you get a BAD
>> prediction for ASP26, who has an experimental pKa of 9.9
>>
>> Now, you KNOW the experiment, and figure out that the problem is that you
>> are using a conformation, or did MD with the ASP deprotonated, which, if
>> the pKa is 9.9, would be wrong, agree?
>>
>> So, they run MD with ASP26 protonated and deprotonated, and indeed, when
>> they use the conformations from the MD run protonated, they get a MUCH
>> better prediction than before.
>>
>> What is the problem here ? That you know you are wrong ONLY if you have the
>> experimental data, which defeats the purpose right ? One would like to have
>> a method that does not need the experiment every time.
>>
>> Second, clearly the pkA prediction depends on the conformers, which depends
>> on the protonation state used in the MD. This is a catch-22 problem !. You
>> do not know the protonation state to run MD until you know the pKa, but
>> cannot trust the pKa until you run the MD at the correct protonation state.
>>
>>
>> That is why I said to be careful. In your advise to the person asking, you
>> said to run MD for 50 ns, then do pKa-ANI, but hopefully you now see the
>> issue, right?
>>
>> What protonation state would you use to run that MD ? For most residues you
>> can just guess, but for the interesting ones, you cannot.
>>
>> Hence cpHMD (which I agree, is a pain to set up!). It couples conformations
>> and protonation states directly, and samples the joint energy surface, so
>> you get all the information together, as you should.
>>
>> BTW: This discussion has nothing to do with ANI per-se. It applies to ANY
>> method that predicts pKa from single structures, even if averaged over MD.
>>
>>
>>
>> On Wed, Feb 21, 2024 at 10:48 PM Dulal Mondal via AMBER <amber.ambermd.org
>> wrote:
>>
>>> Thank you. I understood my mistake.
>>>
>>> On Thu, Feb 22, 2024 at 11:24 AM Dulal Mondal <
>>> babunmondal.chem.kgpian.iitkgp.ac.in> wrote:
>>>
>>>> Thank you for your reply. I tried the grep command on my files and this
>>> is
>>>> the output I got:
>>>> "grep -m1 'Solvent pH:' ../001/cpout.rep.001 ../002/cpout.rep.001
>>>> ../003/cpout.rep.001 ../004/cpout.rep.001 ../005/cpout.rep.001
>>>> ../006/cpout.rep.001 ../007/cpout.rep.001 ../008/cpout.rep.001
>>>> ../009/cpout.rep.001 ../010/cpout.rep.001
>>>> ../001/cpout.rep.001:Solvent pH: 6.40000
>>>> ../002/cpout.rep.001:Solvent pH: 5.70000
>>>> ../003/cpout.rep.001:Solvent pH: 5.60000
>>>> ../004/cpout.rep.001:Solvent pH: 5.90000
>>>> ../005/cpout.rep.001:Solvent pH: 5.80000
>>>> ../006/cpout.rep.001:Solvent pH: 6.10000
>>>> ../007/cpout.rep.001:Solvent pH: 6.00000
>>>> ../008/cpout.rep.001:Solvent pH: 6.30000
>>>> ../009/cpout.rep.001:Solvent pH: 6.20000
>>>> ../010/cpout.rep.001:Solvent pH: 5.90000"
>>>>
>>>> I can see two files showing the same pH values of 5.9, although I am
>>>> unable to understand a few things over here. How exactly is the grep
>>>> command allocating a specific pH value to a cpout file?
>>>>
>>>> On Thu, Feb 22, 2024 at 2:12 AM He, Amy <he.1768.buckeyemail.osu.edu>
>>>> wrote:
>>>>
>>>>> Hi Dulal,
>>>>>
>>>>>
>>>>>
>>>>> A quick way to find two replicas with the same pH is to print the
>>> initial
>>>>> “Solvent pH” in the individual cpout files, something like this:
>>>>>
>>>>> *grep* *-m1* *'Solvent pH:'* ../001/cpout.rep.001 ../002/cpout.rep.001
>>>>> ../003/cpout.rep.001 ../004/cpout.rep.001 ../005/cpout.rep.001
>>>>> ../006/cpout.rep.001 ../007/cpout.rep.001 ../008/cpout.rep.001
>>>>> ../009/cpout.rep.001 ../010/cpout.rep.001*
>>>>>
>>>>>
>>>>>
>>>>> cphstats can’t handle two replicas with the same pH. With these, the
>>>>> sizes of samples at each pH will become unequal, and that can change
>> the
>>>>> error estimates of the pKa fitting, too.. although in my opinion it’s
>>> not
>>>>> wrong to have replicas this close (it’s just like the extreme case of
>>>>> unequally spaced replicas)
>>>>>
>>>>>
>>>>>
>>>>> But if this is not something you planned, make sure to check the
>> inputs
>>>>> and outputs. Do you have multiple files that match
>>> ../010/cpout.rep.001*?
>>>>> Also, cphstats tries to make sure each replica has a unique pH, but it
>>>>> can’t tell whether the outputs were from the same replica exchange
>>>>> calculation or not. So we need to make sure these are correct
>>>>>
>>>>>
>>>>>
>>>>> Hope this is mildly helpful!
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>>
>>>>> Amy He
>>>>>
>>>>> Hadad Lab @ OSU
>>>>>
>>>>> He.1768.osu.edu
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> *From: *Dulal Mondal via AMBER <amber.ambermd.org>
>>>>> *Date: *Wednesday, February 21, 2024 at 1:46 AM
>>>>> *To: *Adrian Roitberg <roitberg.ufl.edu>, AMBER Mailing List <
>>>>> amber.ambermd.org>
>>>>> *Subject: *Re: [AMBER] Constant pH remd
>>>>>
>>>>>
>>>>>
>>>>> Not clear sir.
>>>>>
>>>>> On Tue, 20 Feb, 2024, 11:38 pm Adrian Roitberg via AMBER, <
>>>>> amber.ambermd.org>
>>>>> wrote:
>>>>>
>>>>>> the error is clear, right ?
>>>>>>
>>>>>>
>>>>>> Error: Same pH (5.9) found twice!
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 2/20/24 5:16 AM, Dulal Mondal via AMBER wrote:
>>>>>>> [External Email]
>>>>>>>
>>>>>>> Dear Experts,
>>>>>>> I run the constant pH remd with 10 replicas from pH 5.5 to 6.4
>> with
>>>>> 0.1
>>>>>> pH
>>>>>>> gap. I type the following command
>>>>>>> *cphstats --fix-remd prod001 ../001/cpout.rep.001
>>> ../002/cpout.rep.001
>>>>>>> ../003/cpout.rep.001 ../004/cpout.rep.001 ../005/cpout.rep.001
>>>>>>> ../006/cpout.rep.001 ../007/cpout.rep.001 ../008/cpout.rep.001
>>>>>>> ../009/cpout.rep.001 ../010/cpout.rep.001*
>>>>>>> But it show the following error message.
>>>>>>> Error: Same pH (5.9) found twice!
>>>>>>> Thanking You
>>>>>>> --
>>>>>>> *With regards,*
>>>>>>> *Dulal Mondal,*
>>>>>>> *Research Scholar,*
>>>>>>> *Department of Chemistry,*
>>>>>>> *IIT Kharagpur, Kharagpur 721302.*
>>>>>>> _______________________________________________
>>>>>>> AMBER mailing list
>>>>>>> AMBER.ambermd.org
>>>>>>>
>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>> <
>> http:/lists.ambermd.org/mailman/listinfo/amber
>>>>>> --
>>>>>> Dr. Adrian E. Roitberg
>>>>>> V.T. and Louise Jackson Professor in Chemistry
>>>>>> Department of Chemistry
>>>>>> University of Florida
>>>>>> roitberg.ufl.edu
>>>>>> 352-392-6972
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> AMBER mailing list
>>>>>> AMBER.ambermd.org
>>>>>>
>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>> <
>> http:/lists.ambermd.org/mailman/listinfo/amber
>>>>> _______________________________________________
>>>>> AMBER mailing list
>>>>> AMBER.ambermd.org
>>>>>
>>>>>
>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>> <
>> http:/lists.ambermd.org/mailman/listinfo/amber
>>>>
>>>> --
>>>> *With regards,*
>>>> *Dulal Mondal,*
>>>> *Research Scholar,*
>>>> *Department of Chemistry,*
>>>> *IIT Kharagpur, Kharagpur 721302.*
>>>>
>>>
>>> --
>>> *With regards,*
>>> *Dulal Mondal,*
>>> *Research Scholar,*
>>> *Department of Chemistry,*
>>> *IIT Kharagpur, Kharagpur 721302.*
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
> --
> *Dr. Thomas Steinbrecher*
> Principal Scientist CADD
>
> Roche Pharma Research and Early Development
> Roche Innovation Center Basel
> F. Hoffmann-La Roche Ltd
> Bldg. 092/3.92
> Grenzacherstrasse 124
> 4070 Basel
> Switzerland
>
> Phone +41 61 682 1319
> mailto: thomas.steinbrecher.roche.com
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
Dr. Adrian E. Roitberg
V.T. and Louise Jackson Professor in Chemistry
Department of Chemistry
University of Florida
roitberg.ufl.edu
352-392-6972
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Feb 23 2024 - 10:30:02 PST
