Re: [AMBER] Modes associated to the principal components?

From: Jason Swails <>
Date: Tue, 22 Nov 2016 12:26:31 -0500

On Tue, Nov 22, 2016 at 9:54 AM, The Cromicus Productions <> wrote:

> Hi everyone!
> I'm able to compute the principal components of a given trajectory using
> PCAsuite.
> Then, I can convert them into pdb files and watch the motions.
> PCAsuite also gives me the eigenvalues, so I know which are the most
> important.

> Now, if I understand correctly, the principal components are some "chosen"
> normal modes which summarize the information of the trajectory, but how do
> I know which normal mode is associated to each component I get? In
> particular, I would like to know what is the associated vibration frequency
> of each of the components.
​PCA and normal modes are very different. The "modes" in PCA are
eigenvectors of the covariance matrix. The modes in NMA are eigenvectors
of the Hessian matrix. They *can* contain similar information, but there
is not a 1:1 correlation between PCA modes and normal modes.​ There are
large operational differences, too -- for instance, PCA requires an
ensemble of 'snapshots' (a rather *large* ensemble), whereas NMA requires a
single *minimized* structure.

Using PCA (more specifically using the mass-weighted covariance matrix
rather than the standard covariance matrix) as approximations of
vibrational modes is referred to as the "quasi-harmonic approximation". I
would suggest starting there and scanning the literature.


Jason M. Swails
AMBER mailing list
Received on Tue Nov 22 2016 - 09:30:03 PST
Custom Search