Re: merging abf trajectories with different number of colvars

From: Souvik Sinha (souvik.sinha893_at_gmail.com)
Date: Thu Jun 21 2018 - 10:05:30 CDT

Hi,

The 3rd and 4th column in your *.grad are the projected gradients along
your respective reactions coordinates (i.e. your colvars) which you would
eventually integrate to get your 2D PMF using abf_integrate routine. You
can use this 2D landscape to get your 1D Projection of the PMF along any of
the colvars of your choice. I think that is what Jerome meant in his
reply.

There are many literature where such an approach has been taken to
integrate multidimensional energy landscape to get 1D PMF along a specific
RC. If you are looking for just the equation, you can have a look at this "
https://pubs.acs.org/doi/abs/10.1021/ct5002076".

Regards,
Souvik

On Thu, Jun 21, 2018 at 7:49 PM, Olya Kravchenko <ovkrav_at_gmail.com> wrote:

> Hi Jerome,
>
> thank you, I understand and trying my best, this list is my only source to
> discuss technical questions at this moment.
>
> My two main questions are 1) what is the difference between column 3 and
> column 4 in the *.grad output file produced after running abf with two
> colvars (I did not find a description in the manual for the output format
> for more than one colvar - I am not sure whether I interpret it correctly).
>
> and 2) whether abf_integrate routine can be used in this case (according
> to the manual - yes, but you haven't mentioned it in your first reply and
> started to doubt).
>
> I apologize for too many questions.
>
> Olga
>
>
> On Thu, Jun 21, 2018 at 6:21 AM Jérôme Hénin <jerome.henin_at_ibpc.fr> wrote:
>
>> Hi,
>>
>> This is a fundamental question on the theory, not the software. I can't
>> do a stat mech class on this mailing list. I can recommend Dan Zuckerman's
>> "Statistical Physics of Biomolecules: An Introduction".
>>
>> Best,
>> Jerome
>>
>> On 20 June 2018 at 20:18, Olya Kravchenko <ovkrav_at_gmail.com> wrote:
>>
>>> Hi Jerome,
>>>
>>> thank you for your answer. I am still stuck on the same problem.
>>>
>>> Do I understand it correctly that essentially I have to manually extract
>>> the data and make a 1d data file (in the same format as *.pmf produced with
>>> one colvar) in order to merge them with the rest of my windows? Should I
>>> use abf_integrate (is that what you mean by "integrating conditional
>>> PMFs"?) and then add this data to the plot produced by merging the rest of
>>> the windows? Or should I extract the data from original *.grad and *.count
>>> files without using abf_integrate routine?
>>>
>>> I have two colvars, distance and distanceXY.
>>>
>>> I tried utilizing abf_integrate program (it works and produces *.dev
>>> *.est *.histo and *.pmf files), but I don't understand how to interpret the
>>> output in the resulting *.pmf file and how to extract the data to combine
>>> with with the rest of my plot.
>>>
>>> For example, here is a partial output from my *.pmf file (from
>>> abf_integrate):
>>>
>>> 64.05 0.05 0
>>> 64.05 0.15 1.065
>>> 64.05 0.25 0.97875
>>> 64.05 0.35 1.26875
>>> 64.05 0.45 1.10063
>>> 64.05 0.55 1.24812
>>> 64.05 0.65 1.36687
>>> 64.05 0.75 1.62375
>>> ....
>>> ....
>>>
>>> 64.15 0.05 0.29125
>>> 64.15 0.15 1.26688
>>> 64.15 0.25 1.205
>>> 64.15 0.35 1.39375
>>> 64.15 0.45 1.47813
>>> 64.15 0.55 1.69625
>>> ....
>>> ....
>>>
>>> ...and so on until the first column reaches 70 (which defines the upper
>>> border of the "distance" in my abf window). The second column, as I
>>> understand, represents the values for "distanceXY" (0-4). The third column
>>> is free energy, correct? Then -- how do I choose one value that corresponds
>>> to a particular point of the distance colvar for 1D profile? (i.e. what is
>>> the value of PMF that corresponds to, say, 64.05)
>>>
>>> If I don't have to use abf_integrate and take the data from original
>>> *.grad and *.count files, how do I pick the values in the .grad file?
>>>
>>> My grad file looks approximately like this:
>>>
>>> 64.05 0.05 2.63212 18.2997
>>> 64.05 0.15 1.53358 0.88108
>>> 64.05 0.25 1.19012 5.24787
>>> 64.05 0.35 1.51466 -0.261045
>>> ...
>>> (and so on)
>>>
>>> I don't quite understand what are the last two columns and how to pick
>>> the right values there.
>>>
>>> I would like to understand this process and would really appreciate help!
>>>
>>> Thank you,
>>>
>>> olga
>>>
>>>
>>>
>>>
>>>
>>> On Mon, Jun 4, 2018 at 8:22 AM Jérôme Hénin <jerome.henin_at_ibpc.fr>
>>> wrote:
>>>
>>>> Hi Olga,
>>>>
>>>> The only thing ABF/colvars does is merge data for the same set of
>>>> variables. If you ran ABF in 1d in one case, and 2d in the other case,
>>>> you cannot merge them into a 2d PMF by any simple method. What you can do
>>>> is "marginalize" the 2d dataset to obtain a 1d pmf that can be merged with
>>>> the other one, but that would have to be done manually. Essentially that
>>>> means integrating "conditional PMFs" over one variable, then using that as
>>>> weights to compute the free energy gadient in the other direction as a
>>>> conditional average.
>>>>
>>>> Best,
>>>> Jerome
>>>>
>>>> On 31 May 2018 at 18:13, Olya Kravchenko <ovkrav_at_gmail.com> wrote:
>>>>
>>>>> Hi all,
>>>>>
>>>>> I am trying to merge data from ABF runs, I have several trajectories
>>>>> with distance colvar only and several with distance+distanceXY. I am
>>>>> getting the following error:
>>>>>
>>>>> colvars: Reading sample count from abf2.count and gradient from
>>>>> abf2.grad
>>>>> colvars: Warning: reading from different grid definition (colvar 1);
>>>>> remapping data$
>>>>> colvars: Warning: reading from different grid definition (colvar 1);
>>>>> remapping data$
>>>>> colvars: Reading sample count from abf3.count and gradient from
>>>>> abf3.grad
>>>>> colvars: Error reading grid: wrong number of collective variables.
>>>>> FATAL ERROR: Error in the collective variables module: exiting.
>>>>>
>>>>> What is the best way to put PMF curve together in such case? Should I
>>>>> just extract the data columns that I need or is there a modification to
>>>>> make in merge input files that would account for different number of
>>>>> colvars?
>>>>>
>>>>> Thanks in advance!
>>>>>
>>>>> olga
>>>>>
>>>>>
>>>>
>>

-- 
Souvik Sinha
Research Fellow
Bioinformatics Centre
Bose Institute, India
Contact: 033 25693275

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:21:11 CST