[gmx-users] PCA problems

[Matthias Ernst]

In addition to what Tsjerk said, may I point out that a publication by a
colleague of mine treats exactly these problems that arise when trying
to fit cartesian structures:
http://scitation.aip.org/content/aip/journal/jcp/141/1/10.1063/1.4885338

Regards,
Matthias


On 06/08/2016 06:10 PM, Tsjerk Wassenaar wrote:
> Hi James,
> 
> That's silly! Ambiguous means that the same structure can have multiple
> solutions in a fit. The fit to a single reference structure (with more than
> three atoms) is never ambiguous. Can never, by definition!
> 
> Now if you have two reference structures at hand, and they have (quite)
> different structures, then fitting on one may give a different ensemble
> from fitting on the other. The fit is not consistent, and the inconsistency
> is worse for flexible molecules. Different ensembles will mean different
> correlations, thus giving different principal components.
> 
> Progressive fitting does not solve the problem. In fact, progressive
> fitting _is_ ambiguous. Let's say we have a series of conformations ABCAC.
> Then we  fit C once to B, which was fitted to A, and later we have C fitted
> to A, which was fitted to the previous C. Note that in practice the
> situation will be much worse as we can approach a certain configuration
> from many sides. Using B as reference will yield a fit that is different
> from using A as reference, so the structure C will have two different
> orientations in the resulting ensemble. Hence, the fit is ambiguous.
> 
> For structured proteins, the difference will not matter much. However, in
> long trajectories there may be an added contribution (drift) of the
> orientation.
> 
> Hope it helps,
> 
> Tsjerk
> 
> On Wed, Jun 8, 2016 at 5:33 PM, <jkrieger at mrc-lmb.cam.ac.uk> wrote:
> 
>> Thanks Tsjerk,
>>
>> Isn't the progressive fit supposed to rotate everything back into the same
>> orientation without having to worry about inferring that orientation from
>> a reference structure that doesn't align well? Each configuration should
>> in theory align well to its predecessor all the way back to the starting
>> structure (which is what I'd usually take as a reference anyway).
>>
>> The original note I was thinking of says as follows:
>>
>> '''Before a PCA, all structures should be superimposed onto a common
>> reference
>> structure. This can be problematic for very flexible systems such as
>> peptides,
>> where the fit may be ambiguous, leading to artificial structural
>> transitions. In
>> certain cases, such problems may be alleviated by using a progressive fit,
>> where
>> each structure is superimposed onto the previous one. It is also important
>> to note
>> that when results of different PCAs are to be compared with each other,
>> then
>> each individual PCA should be based on the same reference structure used
>> for
>> superposition.'''
>>
>> Please could you explain further what it is I have misunderstood.
>>
>> Also would you say a progressive fit is a bad idea for more structured
>> proteins?
>>
>> Many thanks
>> James
>>
>>> Hi James,
>>>
>>> 'Spurious alignment' is the dependence of the resulting ensemble on the
>>> reference structure. Unfortunately, that's not solved by a progressive
>>> fit.
>>> Rather, in a progressive fit, the same configuration can have multiple
>>> orientations, based on the previous structures, which is also problematic
>>> when you're trying to understand spatial correlations between atoms
>> within
>>> their reference frame.
>>>
>>> Cheers,
>>>
>>> Tsjerk
>>>
>>> On Wed, Jun 8, 2016 at 9:27 AM, <jkrieger at mrc-lmb.cam.ac.uk> wrote:
>>>
>>>> Dear Teresa,
>>>>
>>>> That sounds like a periodic boundary issue to me. It could be fixed by
>>>> using a tpr instead of a gro as the gmx covar manual says "All
>>>> structures
>>>> are fitted to the structure in the structure file. When this is not a
>>>> run
>>>> input file periodicity will not be taken into account." Alternatively if
>>>> you don't have a tpr you could use gmx trjconv first with -pbc whole or
>>>> -pbc nojump.
>>>>
>>>> I also remember reading (I think it was in the Hayward and de Groot
>>>> review
>>>> 2008) that fitting peptides to a reference structure can cause spurious
>>>> alignments. I don't know if this is also related to what you're seeing
>>>> but
>>>> it might be worth using gmx trjconv again with-fit progressive then use
>>>> -nofit in gmx covar.
>>>>
>>>> Best wishes
>>>> James
>>>>
>>>>> Dear GROMACS community
>>>>>
>>>>> I am trying to complete a PCA analysis of my peptide adsorbed to a
>>>>> surface. However when I use :
>>>>>
>>>>> gmx covar -s trajectory.gro  -f md_golp_vacuo.xtc
>>>>>
>>>>> and select the protein for both the least squares fit and covariance
>>>>> calculation, followed by
>>>>>
>>>>>
>>>>> gmx anaeig -s trajectory.gro -f md_golp_vacuo.trr -filt filter1.gro
>>>>> -first 1 -last 1 -skip 100
>>>>>
>>>>> and I select the peptide for the least squares and covariance
>>>>> calculation
>>>>>
>>>>> My peptide is now broken up into pieces. Is this right?
>>>>>
>>>>>
>>>>>
>>>>> Best
>>>>> Teresa
>>>>> --
>>>>> 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
>>>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>>>> send
>>>>> a mail to gmx-users-request at gromacs.org.
>>>>>
>>>>
>>>> --
>>>> 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
>>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>>>> send a mail to gmx-users-request at gromacs.org.
>>>>
>>>
>>>
>>>
>>> --
>>> Tsjerk A. Wassenaar, Ph.D.
>>> --
>>> 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
>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send
>>> a mail to gmx-users-request at gromacs.org.
>>>
>>
>>
>> --
>> 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
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.
>>
> 
> 
> 


-- 
Progress... http://www.phdcomics.com/comics/archive.php?comicid=1611