What is "Relative Mean Square Deviation (rmsd)" in Swiss-Model???

EdiSon

unread,

Sep 2, 2001, 7:50:03 AM9/2/01

to

Dear All,

RMSD (Relative Mean Square Deviation)is used to assess the Reliability
of Swiss-Model for proteins. Can anyone explain the meaning of RMSD in
plain simple English?? Say for example:
"Percent models with rmsd lower than 2 Å"... what the heck does this
mean ~__~?

As in :

http://www.expasy.org/swissmod/SWISS-MODEL.html

many thanks,
-E...

Barry in Maryland

unread,

Sep 5, 2001, 7:45:58 AM9/5/01

to

molub...@hotmail.com (EdiSon) wrote in message news:<5c8d8266.0109...@posting.google.com>...

Maybe they mean *root mean square*?
Barry in Maryland

Marvin Margoshes

unread,

Sep 5, 2001, 2:50:25 PM9/5/01

to

"Barry in Maryland" <brwil...@my-deja.com> wrote in message
news:fe83f2ac.01090...@posting.google.com...

Sounds right, but "relative" probably means it is expressed as a fraction
(or a percentage of the average value of the parameter.

Barry in Maryland

unread,

Sep 6, 2001, 6:58:56 AM9/6/01

to

> > Maybe they mean *root mean square*?
> > Barry in Maryland
>
> Sounds right, but "relative" probably means it is expressed as a fraction
> (or a percentage of the average value of the parameter.

I suspect you're probably right, but after going out to the link that
was in the original post, I decided that it was unclear. Usage of
uncertainty terms can get a little fuzzy, and this one is a good
example. Since this is a Swiss citation, I'm surprised they don't use
the ISO terminologies specified in the Eurachem/CITAC guide, which is
a Swiss organization. The guide is on the Web at
http://www.measurementuncertainty.org

Then again, I guess I'm not suprised.

Barry in Maryland

EdiSon

unread,

Sep 7, 2001, 5:31:41 AM9/7/01

to

Dear Barry and Marvin ^^"... i've received the reply below via
e-mail... hopefully u might find it interesting, too .

cheers, E...
---------------------------------------------------------------------------
Root-mean square deviation for protein-protein comparisons is usually
calculated as a score that refelects how far (on average) are the
matching
C-alpha atoms. If you have gaps in protein alignments, the atoms in
the gaps
are, obviously, excluded.

sigma = sqrt(sum[d(n)^2]/N) where d(n) is the individual deviation and
N is
number of atoms compared.

For a general case, see
http://www.its.bldrdoc.gov/fs-1037/dir-031/_4633.htm

The lower your RMSD, the better is the correlation between your
models. The
trick is to applly RMSD only to the atoms that make sense, and exclude
the
residues that are in gaps, or (sometimes) widely different loops.

A.G.E.

Marvin Margoshes

unread,

Sep 7, 2001, 3:01:37 PM9/7/01

to

"EdiSon" <molub...@hotmail.com> wrote in message
news:5c8d8266.01090...@posting.google.com...

The explanation confused me, since I don't know what is meant by "how far
(on average) are the matching C-alpha atoms." How far from WHAT? The URL
helped somewhat, but not fully because the meanings of the parameters aren't
defined there. It still looks to me like the statistical parameter,
relative standard deviation.

Artem Evdokimov

unread,

Sep 7, 2001, 9:07:43 PM9/7/01

to

Root Mean Square Deviation, not relative.

> The explanation confused me, since I don't know what is meant by "how far
> (on average) are the matching C-alpha atoms." How far from WHAT?

Fro each other, obviously. Atoms are compared in pairs, the pairwise
relation is established either via series comparison of sequences or via
least squares fit optimizaiton. We're talking about a specific case of RMSD
used to compare two (or more) protein structures.

The URL is just a standard definition of an RMSD for any general case.

Take care,

A.G.E.