Project

General

Profile

Feature #438

New tool: g_correl

Added by Alexey Shvetsov over 9 years ago. Updated over 5 years ago.

Status:
In Progress
Priority:
Low
Category:
analysis tools
Target version:
Difficulty:
uncategorized
Close

Description

I write small utility to compute Pearson's product-moment coefficients for groups of atoms from md trajectory. This tool reads trajectory and constructs correlation matrix of atomic displacements between 2 index groups.

0001-Add-new-tool-g_correlation.-This-tool-computes-Pears.patch (14.3 KB) 0001-Add-new-tool-g_correlation.-This-tool-computes-Pears.patch patch againts git master Alexey Shvetsov, 06/15/2010 01:31 AM
0001-Add-new-tool-g_correl.patch (14.8 KB) 0001-Add-new-tool-g_correl.patch new patch againts git master Alexey Shvetsov, 06/15/2010 04:49 PM
0001-Add-g_correl-tool.patch (18.3 KB) 0001-Add-g_correl-tool.patch Rebased patch Alexey Shvetsov, 08/11/2010 03:32 PM

History

#1 Updated by Alexey Shvetsov over 9 years ago

Created an attachment (id=470)
patch againts git master

Patch against git master branch
use git am to apply it =)

#2 Updated by Rossen Apostolov over 9 years ago

g_covar already calculates the covariance matrix, which can be easily extended to print the correlations too.

Also, there is already a tool named g_correlation, which uses mutual information to estimate the coefficients:

http://www.mpibpc.mpg.de/home/grubmueller/projects/MethodAdvancements/GeneralizedCorrelations/index.html

#3 Updated by Alexey Shvetsov over 9 years ago

(In reply to comment #2)

g_covar already calculates the covariance matrix, which can be easily extended
to print the correlations too.

Also, there is already a tool named g_correlation, which uses mutual
information to estimate the coefficients:

http://www.mpibpc.mpg.de/home/grubmueller/projects/MethodAdvancements/GeneralizedCorrelations/index.html

Yes. g_covar calculates covariance matrix but this matrix is only for group.
While my utility can compute cross-correlations (correlation between 2 groups)
Also doesnt works with current gromacs (it only works with gromacs 3.x)

[1] http://www.mpibpc.mpg.de/home/grubmueller/projects/MethodAdvancements/GeneralizedCorrelations/index.html

#4 Updated by Alexey Shvetsov over 9 years ago

Hmm. Seems name of my utility is a bit confusing. So i should change it.

Also it computes a little bit different thing comparing to [1]. And as i mentioned earlyer it can compute correlations between 2 different index group.

Correlation coefficient defined here as following

C_ij = <(R_i-<R_i>)*(R_j-<R_j>)>/sqrt(<(R_i-<R_i>)^2>*<(R_j-<R_j>)^2>)

where R_i, R_j are position vectors for atoms i and j respectevely

so C_ij can be between -1 (for anticorrelated motions) and +1 (for correlated motions)

[1]
http://www.mpibpc.mpg.de/home/grubmueller/projects/MethodAdvancements/GeneralizedCorrelations/index.html

#5 Updated by Alexey Shvetsov over 9 years ago

Created an attachment (id=473)
new patch againts git master

Reformated patch.
Utility was renamed to g_correl to not confuse people.
Also i addeddescription to git commit message about how it works

#6 Updated by Rossen Apostolov over 9 years ago

I still don't understand the need for this utility. The correlations between 2 groups of atoms will be simply a sub-matrix of the full system matrix. Or am I missing something? And an extra option to g_covar will be a simpler solution.

#7 Updated by Alexey Shvetsov over 9 years ago

(In reply to comment #6)

I still don't understand the need for this utility. The correlations between 2
groups of atoms will be simply a sub-matrix of the full system matrix. Or am I
missing something? And an extra option to g_covar will be a simpler solution.

Well basicaly its the same as for pair g_rms/g_rmsf. g_correl is very simple tool to compute correlation matrix of needed parts of system e.g. between 2 monomers from multiprotein complex.

#8 Updated by Erik Lindahl about 9 years ago

Hi,

We won't be able to get a new correlation tool working for 4.5, but we should work on updating & integrating the Grubmüller tool (or rather, get Gerrit or Bert to do it), and then we'll also make sure that it supports correlations between disjoint groups.

#9 Updated by Alexey Shvetsov about 9 years ago

(In reply to comment #8)

Hi,

We won't be able to get a new correlation tool working for 4.5, but we should
work on updating & integrating the Grubmüller tool (or rather, get Gerrit or
Bert to do it), and then we'll also make sure that it supports correlations
between disjoint groups.

Hmm. I can make partial rewrite of Grubmüller g_correlation tool to make it works with current git master.

#10 Updated by Alexey Shvetsov about 9 years ago

Created an attachment (id=511)
Rebased patch

Rebased version of tool that computes Peerason correlations

#11 Updated by Szilárd Páll over 8 years ago

  • Assignee deleted (Erik Lindahl)
  • Target version deleted (CVS)

#12 Updated by Teemu Murtola over 6 years ago

  • Tracker changed from Bug to Feature

#13 Updated by Teemu Murtola over 6 years ago

  • Target version set to 5.0

Set the target version to 5.0, as it was already noted in some comments that this functionality could be included in one of the tools. So for 5.0, as part of the larger work in #665, we should either identify the best solution for including this functionality, or close this issue (or postpone if no one cares enough to do the work in that timeframe).

#14 Updated by Alexey Shvetsov over 6 years ago

Well I can rewrite it using new framework for 5.0

#15 Updated by Rossen Apostolov over 5 years ago

Alexey, are you still planning to port it?

#16 Updated by Alexey Shvetsov over 5 years ago

Yep. I'll do it after I finish rewrite of g_sans using new framework, may be in addition i'll rewrite g_covar since it uses very similar ideas.

#17 Updated by Rossen Apostolov over 5 years ago

  • Status changed from New to In Progress
  • Assignee set to Alexey Shvetsov

OK, thanks!

#18 Updated by Erik Lindahl over 5 years ago

  • Target version changed from 5.0 to future

Also available in: Atom PDF