[gmx-users] dPCA analysis

[Dongsheng Zhang]

Dear All,

According to David's instruction at
 http://www.gromacs.org/pipermail/gmx-users/2005-October/017512.html

gromax 3.3 can be used to do dPCA (dihedrals PCA) analysis by
1. g_angle -or to get angle.trr
2. g_covar -f angle.trr -s *.tpr to get eigenvec.trr
3. g_anaeig -f angle.trr -v eigenvec.trr -proj to get the projection of
angle.trr on eigenvec.trr

However, when I tried it in the way I mentioned above, I did not get
what I expected. Therefore, I download Dr. Mu's code from
http://www.gromacs.org/contributions/index.php

and did the analysis again following Mu's recipe. I got a totally
different result. 

After my investigation, I have a question regarding to the dPCA method
in gromacs 3.3. In the conventional PCA, we need to remove the
translational and rotational motion before generating the covariance
matrix. That is what I think g_covar is actually doing. 

In dPCA, we should not do any modification of coordinates in angle.trr.
I mean, no translation and rotation removal. This should be so because
the dihedrals (or cos and sin) are internal coordinates.
Then, I think we can not use g_covar directly to analyze angle.trr,
rather we need to modify g_covar.

Has anyone used the gmx3.3 to do dPCA analysis? If so, please give me
your comments on my thoughts. Thank you in advance!



Best regards!


Dongsheng