Re: [AMBER] Eigenvalues from PCA in cpptraj

From: Daniel Roe <daniel.r.roe.gmail.com>
Date: Wed, 1 Aug 2018 12:59:16 -0400

Did you use a mass-weighted covariance matrix? If so, be aware the
eigenvalues get converted to frequencies which are inversely
proportional to the raw eigenvalues. The reasons why this is done are
historical and have to do with doing subsequent thermodynamic
analysis. There should probably be an option to either prevent this or
correct for it in 'modes' - I'll open up a GitHub issue for it to
remind me.

In the meantime, you can repeat your analysis without mass-weighting
(i.e. use a normal covariance matrix) and you should get output that
you expect.

Hope this helps,

-Dan

On Wed, Aug 1, 2018 at 12:49 PM, Adam Jo Elatico <ajelatico.up.edu.ph> wrote:
> Dear Daniel,
>
> Thanks for the response :)
>
> I'm really confused as to why the fractional contribution of mode 1 is the
> least of all the modes analyzed. Why is this so?
>
> On Thursday, August 2, 2018, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>
>> Hi,
>>
>> By convention, cpptraj orders eigenvectors (and eigenvalues) from most
>> significant to least significant. In "evecs" file output, the
>> eigenvalue is printed right before its associated eigenvector.
>>
>> Hope this helps,
>>
>> -Dan
>>
>> On Wed, Aug 1, 2018 at 7:39 AM, Adam Jo Elatico <ajelatico.up.edu.ph>
>> wrote:
>> > Hi all,
>> >
>> > I'm trying to perform principal component analysis on my system
>> trajectory
>> > using cpptraj. I've noticed that the eigenvalue for each mode increases
>> > with increasing mode number. Thus, the fractional contribution of each
>> > eigenvalue increases as well.
>> >
>> > What does this mean? Does it mean that mode #1 does not correspond to PC1
>> > (i.e., the PC with the largest contribution to overall motion)? Please do
>> > enlighten me.
>> >
>> > Many thanks!
>> > Adam
>> >
>> >
>> > --
>> >
>> > *Adam Jo J. Elatico*
>> > Science Researcher
>> > University of the Philippines Diliman
>> > Quezon City, Philippines 1101
>> > Phone: +639364081744
>> > _______________________________________________
>> > AMBER mailing list
>> > AMBER.ambermd.org
>> > http://lists.ambermd.org/mailman/listinfo/amber
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>
> --
>
> *Adam Jo J. Elatico*
> Science Researcher
> University of the Philippines Diliman
> Quezon City, Philippines 1101
> Phone: +639364081744
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber

_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 01 2018 - 10:00:02 PDT
Custom Search