HI,

Previously, I searched regarding this topic, and found this elaborated answer by 

Matthias Ernst. He quite elegantly explain the underlying science and method. I am just copying his answer below. Hope it will help.

Best,

dPCA was developed in the group where I'm working now and I have to say that the way it is described in the GROMACS howto (http://www.gromacs.org/Documentation/How-tos/Dihedral_PCA) feels like a dirty hack. It works, though, if you do it correctly. The trick is that you first do the sin- and cos-transformation of the dihedrals, save them to a kind of "dummy" trajectory, create a kind of "dummy" structure just to get the right number of coordinates and then evaluate it using the standard GROMACS tools.

Let me briefly summarize what the tutorial tells you to do (I refer to their numbering) and what is actually menat:

A) choose the dihedral angles you need (1. in tutorial, given by 4 atoms per dihedral) and do the sin- and cos-transformation. This is then saved as a new GROMACS trajectory (called dangle.trr in the tutorial). BUT, the values are saved in triples of variables as the trajectory was initially intended for cartesian coordinates, meaning there can be "empty" coordinates which are 0.0 always if the number of coordinates you use is not divisable by 3.

B) create a dummy index file which just tells g_covar how many atoms you need for storing all values you have. Step 3 says you need to put integer(2*N/3) atoms in the index file. This means: you have N dihedrals and doing the sin- and cos-transformation, you get 2*N coordinates. They are stored in triples of 3 as mentioned above, so you need M=2*N/3 atoms but you need to round up (indicated by the "integer") to make sure you don't drop one or two coordinates. That's why in the example you need to count up to 4 in the index file (10 coordinates after transformation meaning 3,33 atoms => round up to 4)

C) Create also a dummy gro file which also contains M atoms. This can contain any values whatsoever, it just has to be a .gro file with the correct number of atoms. Step 4 in the tutorial does this.

D) To actually perform the PCA, you need to diagonalize the covariance matrix of your coordinates. This is done by g_covar in step 5. You need to provide your dummy gro and index files as well as some options to tell the tool not to do fitting, mass weighting or whatever, which is why there are so many flags. If something goes wrong here, I would assume that there is an error in the files you had to create in the steps before.

E) To evaluate your PCA, you can plot one- or two-dimensional projections of the data onto the eigenvectors of the covariance matrix (the so-called principal components). This is done in steps 6 to 8.

And now to answer your question: I cannot tell you what actually went wrong because I don't know your exact files and input and so I don't know the reason for the segfault (can be many...). Please make sure you understand what you do and that you did it correctly (I hope, the text above can help).

If you are sure you did everything correctly but the errors still persist, please provide us with more information, only then we can help you.

Good luck and success!

Can anyone help me with doubts regarding dihedral PCA? - ResearchGate. Available from: https://www.researchgate.net/post/Can_anyone_help_me_with_doubts_regarding_dihedral_PCA [accessed Apr 19, 2016].

That is a very satisfactory answer..thank u