AMBER: PCA - analysis

From: Ibrahim Moustafa <I.moustafa.psu.edu>
Date: Sun, 01 Jun 2008 14:00:35 -0400

Dear all,

  Iım trying to do PCA on a 20 ns trajectory and Iım new to this type of
analysis. The calculation was done in Amber9 and PTRAJ used to do the
analysis using the following script:
>>
Tarjin trejectoryfile.mdcrd 2001 20000 1 # 18,000 frames are used for the
analysis
Rms first :1-461.CA
Matrix covar name pcacovar :1-461.CA out pcacovar18.mat
Analyze matrix pcacovar out eigenvec18.dat name eigenvect vecs 1383
Analyze modes fluct stack eigenvect out rmsfluct_PC1.dat beg 1 end 1
Analyze modes fluct stack eigenvect out rmsfluct_PC2.dat beg 2 end 2
....etc
Analyze modes fluct stack eigenvect out rmsfluct_PC10.dat beg 10 end 10

Analyze modes disp stack eigenvect out rmsfluct_PC1.dat beg 1 end 1
...etc
Analyze modes fluct stack eigenvect out rmsfluct_PC10.dat beg 10 end 10

>>
Iım analyzing the first 10 PCs (willing to look at the 1st 50 PCs) and want
to look at the PCs individually so I can visualize the corresponding
displacements on the structure.

The eigenvect output file has the 1383 eigenvectors and values, below is a
portion of PC1 in the eigenvector output file:

1 218.63728

-0.00750 -0.01603 -0.00507 -0.00535 -0.01348 -0.00057 -0.00379
   -0.01278 0.00106 -0.00213 -0.02327 0.00619 0.00277
-0.02486
    0.01126 0.00597 -0.02036 0.00711 0.00763 -0.02560
0.01738
    0.00825 -0.02390 0.01519 -0.00756 -0.03153 0.01020
-0.00826
   -0.03982 0.01204 0.00333 -0.04672 0.01886 -0.00328
-0.04875
    0.01634 -0.01431 -0.05148 0.00831 -0.02311 -0.03459
0.01062
   -0.03168 -0.03037 0.00984 -0.01174 -0.00791 -0.00527
-0.01109
   -0.02305 -0.02891 -0.00690 0.01773 -0.05663 -0.01479
0.04973
   -0.09230 -0.02303 0.07188 -0.07695 -0.02791 0.09910
-0.08638

  The rmsd analysis of the 1st PC looks like that:
Analysis of modes: RMS FLUCTUATIONS
  Atom no. rmsX rmsY rmsZ rms
         1 0.003 0.006 0.002 0.007
         2 0.002 0.005 0.000 0.006
         3 0.001 0.005 0.000 0.005
         4 0.001 0.009 0.002 0.009
         5 0.001 0.010 0.004 0.011
         6 0.002 0.008 0.003 0.009
         7 0.003 0.010 0.007 0.012
         8 0.003 0.009 0.006 0.011
         9 0.003 0.012 0.004 0.013
        10 0.003 0.015 0.005 0.016


   Iım confused because of the low rmsd values obtained for the 1st
eigenvector, which is supposed to have the largest proportion of the
variance (27% of the total variance).
The values for the 1st 6 PCs are of the same order and getting increase for
higher PCs ranking; for PC10:

Analysis of modes: RMS FLUCTUATIONS
  Atom no. rmsX rmsY rmsZ rms
         1 0.078 0.188 0.038 0.207
         2 0.132 0.170 0.109 0.242
         3 0.004 0.071 0.019 0.074
         4 0.106 0.013 0.184 0.213
         5 0.023 0.086 0.201 0.220
         6 0.047 0.123 0.159 0.207
         7 0.069 0.165 0.138 0.226

 It is not clear to me why the rmsd from the first PCs (1 ­ 6, accounting
for 67% of the total variance) has such a low value? Isnıt it supposed to
correspond to the largest motion?
Is that normal for the 1st PCs analyzed in PTRAJ to have such low values?

PLZ, correct me if the above protocol used for PCA is not the proper way to
carry out such analysis. What is the right way to evaluate the PCs and to
get insights into the observed motion of the trajectory?.

  Many thanks in advance for your help and comments.

  Best regards,
 Ibrahim





-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
      to majordomo.scripps.edu
Received on Wed Jun 04 2008 - 06:07:12 PDT
Custom Search