[gmx-users] anomalous eigenvector in PCA (Principal Component Analysis)

[Nima Soltani]

Justin Lemkul <jalemkul at ...> writes:

> 
> 
> On 12/27/15 6:48 PM, Nima Soltani wrote:
> > Hi
> > I have performed principal component analysis by using g_covar module of
> > Gromacs 5.0  (Peptide's backbone atoms were selected both for least square
> > fit and covariance analysis)
> > then 2 extreme conformations along different eigenvectors derived by
> > "g_anaeig -extr"
> > The resulting structure look likes this :
> > https://app.box.com/s/e6b87ar9j75q227qe6sx1udeyqdszgdl
> > ​Is there any way to correct those artifices that can be observed in the
> > peptide backbone?
> 
> Looks to me like you're just analyzing a highly polymorphic peptide.  Maybe 
> there's nothing even useful to conclude from such analysis.  How long is the 
> simulation?  What do you know about the conformational ensemble you've 
> simulated, and how well it is converged?
> 
> -Justin
> 
Thank you Dear Dr. Lemkul for your reply 
Actually it is a portion of Alzheimer's A-Beta peptide, i.e. simulated in
different environments, each simulation is approximately 100ns long.
In those simulations A-B adsorbs to different surfaces and in it's adsorbate
state has only a few conformationally allowed states. 
not every eigenvector looks as bizarre as this one.
Could it be wrong choice of reference structure that caused this ?