[gmx-users] PCA comparison

[Ramon Crehuet]

Dear all,
We are performing some PCA analysis of several 22ns trajectories of a
protein hexamer at different temperatures (280, 300, 320K). We expected
to see a similar movement decribed by the ~10 lowest PCA's.
their overlap is very poor: Below 0.35, and the diagonal elements are
not remarkably higher than the rest. If we split one of the trajectories
into to 11ns ones and do the same, again the overlap obtained is below
0.44 in all cases and there is no clear diagonal. 11ns is quite a long
trajectory so I expected to see the same type of movement described by
both 11ns fragments. Here are the cosine contents of the whole 22 ns
trajectory and the two fragments. Except for the 1st pc, the rest seem
well sampled: shouldn't they agree?
22ns:
Cosine content  PC1 0.956255
Cosine content  PC2 0.00428106
Cosine content  PC3 0.000162083

1st 11ns:
Cosine content  PC1 0.923261
Cosine content  PC2  0.0217681
Cosine content  PC3 0.0135543

2nd 11ns:
Cosine content  PC1 0.786694
Cosine content  PC2 0.1654
Cosine content  PC3 0.000335589

Are we doing something wrong? Is this the exptected behaviour?
This is what we do to calculate the PCA's (10ns of equilibration removed):
g_covar -f trajectory.xtc -b 10000 -e 21000 -s structure.pdb -o
eigenval-T300-A.xvg -v eigenvec-T300-A.trr -av average-pca-T300-A.pdb
g_covar -f trajectory.xtc -b 21020 -s structure.pdb -o
eigenval-T300-B.xvg -v eigenvec-T300-B.trr -av average-pca-T300-B.pdb

and to compare them:
g_anaeig -inpr T300-A-B.xpm -v eigenvec-T300-A.trr -v2
eigenvec-T300-B.trr -first 1 -last 10

Thanks for your help and suggestions!
Ramon