[gmx-users] applications of principal component analysis

[Ali Hassanali]

Dear Gromacs Users,

I have been working on a protein system for a while and have been using 
principal component analysis (PCA) to analyze the various protein motions 
that occur in a localized region of the protein during a timescale of 
about 100ps. I am interested in quantifying both the nature and timescales 
of the waters within 15 Angstroms of this region on the protein during 
this structural transition as well.  I basically conduct the PCA, the same 
way I did for the protein, except I do it for the waters within 15 A of a 
certain region on the protein. Part of the challenge of doing this for the 
waters is that waters translate by a larger magnitude than protein atoms 
from time t=0 to t=100ps.  So using the waters that were within 15 Ang of 
the protein, I obtained the principal components, and then obtained a 
filtered trajectory along the first principal component to see what the 
water motions that occured were. What I do observe is that both the 
protein principal component and the water principal component undego a 
transition as the protein structural change occurs. The transition in the 
protein principal component makes sense because that tracks the structural 
transition that I am interested in analyzing. The transition in just the 
hydration water component may not be surprising either.

I basically wanted to get a sense from anyone with more experience in the 
technicalities of PCA, as to whether there was any advice/suggestions on 
how to do the PCA for the water surrounding that region. Ideally I would 
like to do it for all waters, but thats a very huge matrix to diagonalize 
and as mentioned earlier waters migrate around the box.

Essentially I am only interested in the collective water motions around a 
certain region of the protein over a period of ~100ps. These water 
motions facilitate protein motions in some way and I am hoping to quantify 
this better.

Ali Hassanali