[gmx-users] Principal Components Analysis in Gromacs

[Tsjerk Wassenaar]

Hi Thomas,

> Thanks for all the clarifications about PCA you make on the mailing list!

Thank you for the appreciation :)

> I have a question about the commandlines you wrote. Why do you use the .tpr
> file with the "-s" flag? Is it because you want to compare the
> mass-wheighted covariance matrices? I use to calculate the covariance
> matrices by giving to g_covar a .pdb file with the "-s" flag and then
> calculate the RMSIP without giving any structure file. I guess no masses are
> used in that covariance analysis, right? Do you recommend using atom masses
> for PCA in general?

Well, I admit that in most cases I don't use mass-weighting myself.
Unless you include hydrogens, it also doesn't matter much, as the
masses are not very different. Only if you want to calculate
frequencies, e.g. to connect to NMA and/or IR spectroscopy you would
really need masses.
If you use a .pdb or .gro file, you don't get mass-weighting. And
you're right that for calculating the RMSIP, and the subspace overlap,
and the martix of inner products, you don't need a structure filel,
but only the eigenvectors, and possibly the eigenvalues.

Cheers,

Tsjerk



-- 
Tsjerk A. Wassenaar, Ph.D.

post-doctoral researcher
Molecular Dynamics Group
* Groningen Institute for Biomolecular Research and Biotechnology
* Zernike Institute for Advanced Materials
University of Groningen
The Netherlands