Hi,
Out of curiosity do you see the same thing happen if you use
non-mass-weighted covariance matrices (i.e. covar instead of mwcovar)?
Note that typically in PCA just plain covariance matrices are used.
Mass-weighted covariance matrices are really only required if you want
to perform quasi-harmonic analysis.
-Dan
On Tue, Nov 25, 2014 at 10:49 AM, Guillaume Roellinger
<roellinger.bio.mx> wrote:
> Dear Amber users,
>
> Using cpptraj, I am trying to analyze two 10ns MDs in one principal component analysis (PCA). Both MDs are closely related to each other: it is the same protein but in two different conformations (conf1 and conf2) with same number of residues. I was interested to include the structures of each conformations just after equilibration as reference structures for the PCA. I autoimaged and removed water and ions from my trajectories and reference structures and generated the corresponding stripped topologies conf1.prmtop, conf2.prmtop for each conformation. Before calculating the mass-weighted matrix, I aligned all frames to an average structure generated from both trajectories (createAverageStructure_cpptraj file).
>
> The problem is that I get two very different results in my evecs file when switching the loading order between the reference structures and trajectories and I have no idea why.
> By loading the reference structure first (PCA_cpptraj_1), I get the following 5 first eigenvalues:
> 0.85663
> 3.57670
> 5.85045
> 7.69400
> 8.09997
> ...(and so on, increasing linearly)
>
> And by loading the MD trajectories first (PCA_cpptraj_2):
> 26.45372
> 2640022.22255
> 2642822.92226
> 2769903.48459
> 2844226.43589
> ...(and so on, continuing with values ~2e6, 3e6)
>
> If I try not to load any reference structures, I get almost the same results as when loading the reference structure first (PCA_cpptraj_1); so I checked the reference structures and they are both fine. I checked also the structures after the RMSD on the average structure with outtraj and in both cases the structures are correctly aligned.
> If necessary, I can provide a 12MB archive containing the input files and shortened MDs trajectories (20 first frames of each) which is sufficient to reproduce this behavior.
>
> I am using cpptraj V14.09 and calling cpptraj via $AMBERHOME/bin/cpptraj -i file_cpptraj
>
> Thanks for your help,
>
> Best regards,
>
> G.R.
>
> Input Files used:
>
> createAverageStructure_cpptraj:
> ------------------------------------------------------------
> parm conf1.prmtop
> parm conf2.prmtop?
> trajin conf1_MDtraj.mdcrd parmindex 0
> trajin conf2_MDtraj.mdcrd parmindex 1?
> rms RMSDfirst first out RMSDfirst.dat .CA
> average averageCA.pdb .CA
> ------------------------------------------------------------
>
>
> PCA_cpptraj_1 (loading reference structures first):
> ------------------------------------------------------------
> parm conf1.prmtop
> parm conf2.prmtop
> trajin conf1_afterEquil.rst7 parmindex 0
> trajin conf2_afterEquil.rst7 parmindex 1
> trajin conf1_MDtraj.mdcrd parmindex 0 # 500 frames
> trajin conf2_MDtraj.mdcrd parmindex 1 # 500 frames
> parm averageCA.prmtop
> reference averageCA.pdb parmindex 2
> rms RMSDaverage reference out RMSDaverage.dat .CA
> matrix mwcovar name mwcvmat out mwcvmat.dat .CA
> diagmatrix mwcvmat out evecs.dat vecs 200 name mymodes
> run
> readdata evecs.dat
> projection modes evecs.dat out project.dat beg 1 end 50 .CA
> ------------------------------------------------------------
>
> PCA_cpptraj_2 (loading MD first):
> ------------------------------------------------------------
> parm conf1.prmtop
> parm conf2.prmtop
> trajin conf1_MDtraj.mdcrd parmindex 0 # 500 frames
> trajin conf2_MDtraj.mdcrd parmindex 1 # 500 frames
> trajin conf1_afterEquil.rst7 parmindex 0
> trajin conf2_afterEquil.rst7 parmindex 1
> parm averageCA.prmtop
> reference averageCA.pdb parmindex 2
> rms RMSDaverage reference out RMSDaverage.dat .CA
> matrix mwcovar name mwcvmat out mwcvmat.dat .CA
> diagmatrix mwcvmat out evecs.dat vecs 200 name mymodes
> run
> readdata evecs.dat
> projection modes evecs.dat out project.dat beg 1 end 50 .CA
> ------------------------------------------------------------
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
-------------------------
Daniel R. Roe, PhD
Department of Medicinal Chemistry
University of Utah
30 South 2000 East, Room 307
Salt Lake City, UT 84112-5820
http://home.chpc.utah.edu/~cheatham/
(801) 587-9652
(801) 585-6208 (Fax)
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Nov 25 2014 - 10:30:02 PST
