[gmx-users] Re: dPCA

[David]

On Thu, 2005-10-13 at 13:36 -0700, Gia Maisuradze wrote:
> David,
> 
> Thank you very much for your reply. As you said, dPCA shows more minima and 
> free energy landscape is more rugged than in "normal" PCA. According to the 
> paper that I mentioned in my previous email, dPCA is more realistic, and 
> Cartesian PCA represents an artifacts due to the mixing of internal and 
> overall motion. On the other hand, there are other opinions, for example 
> Erik Lindahl mentioned that the resulting eigenvectors are no longer 
> orthogonal in the cartesian norm and dPCA is problematic.
> 
> I am going to test dPCA and compare the results to normal PCA. Maybe it can 
> clarify some problems.

We are currently considering some other possibilities as well. A recent
paper by the Levy group used PCA on a set of internal distances (e.g.
Ca-Ca) which has the advantage that you don't need to do a LSQ
superposition, but also that long distances weigh differently than short
distances. A problem in any method is that even if you obtain orthogonal
eigenvectors they may still be coupled. If you find a general solution
for these problems it may be very useful indeed.

> 
> With best wishes,
> 
> Gia
> 
> ----- Original Message ----- 
> From: "David" <spoel at xray.bmc.uu.se>
> To: "Gia Maisuradze" <gia at chem.unr.edu>
> Cc: <gmx-users at gromacs.org>
> Sent: Thursday, October 13, 2005 1:03 PM
> Subject: Re: dPCA
> 
> 
> > On Thu, 2005-10-13 at 11:48 -0700, Gia Maisuradze wrote:
> >> Dear David,
> >>
> >> I have checked the changes in new Gromacs 3.3. It says that dihedral
> >> PCA is supported in g_angle (you are the author). Does it mean that we
> >> can do dihedral principal component analysis for proteins? Is it the
> >> same type of analysis that has been done by Gerhard Stock' group
> >> (PROTEINS, 58, 45, 2005)? If yes, I don't clearly understand how I can
> >> do that. Normal PCA (in cartesian coordinates) can be done by using
> >> g_covar, where we build the covariance matrix and diagonalize it, and
> >> get eigenvectors and eigenvalues. Then analyze them.
> >>
> >> Could you, please, explain how I can do dPCA on Gromacs, if it is
> >> possible.
> > Yes, you use g_angle -oc to produce a new trajectory file containing:
> > cos(phi) sin(phi) cos(psi) sin(psi) etc.
> > This you run through g_covar.
> >
> > According to some people this information is not really useful for
> > protein analysis, as it can show many false minima and it is hard to go
> > back to real space. Barriers you see may be trivial due to overlapping
> > atoms.
> >
> >
> >>
> >> The same question I have asked Erik Lindahl, but he did not know about
> >> dPCA in new Gromacs version.
> >>
> >> Thanks,
> >>
> >> Gia
> > -- 
> > David.
> > ________________________________________________________________________
> > David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group,
> > Dept. of Cell and Molecular Biology, Uppsala University.
> > Husargatan 3, Box 596,          75124 Uppsala, Sweden
> > phone:  46 18 471 4205          fax: 46 18 511 755
> > spoel at xray.bmc.uu.se    spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
> > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >
> > 
-- 
David.
________________________________________________________________________
David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group,
Dept. of Cell and Molecular Biology, Uppsala University.
Husargatan 3, Box 596,          75124 Uppsala, Sweden
phone:  46 18 471 4205          fax: 46 18 511 755
spoel at xray.bmc.uu.se    spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++