Date: Tue, 19 Nov 2002 09:51:53 -0800 (PST)
From: Thomas Cheatham
Subject: Re: limit of RMSD value


> I'd like to know is it any maximum limit for RMSD value? Is it any
> criteria for RMSD value to say our structure is not distorted from
> starting structure? I agree with people who say above 2 Angstrom is very
> distorted structure, but why 2 Angstrom? Why not 1.5 or 0.5 Angstrom?

I do not beleive there is any hard and fast rule to state what is the
"maximum" RMSD for a given system nor what RMSD value is representative of
distortion or good structure. The RMSD value is very crude and sometimes
misleading since a small rotation (say of one subunit of a protein
relative to another or a hinge motion) can lead to large change in RMSd
yet very little overall change in the structure. For most small proteins
or nucleic acid systems (10-20 base pairs), a 2.0 angstrom deviation is
still rather close and thermal fluctuations alone lead to differences in
the 1-1.5 angstrom (all atom) range. What values represent good and
distorted structure depends on the system (and definition of what is
good).

Regarding the maximal RMSd value possible, this also depends on the system
(i.e. how many atoms and the range of possible structures). A maximal
RMSd might be inferred by creating a purely linear structure of your
protein and calculating the RMSd of this to the native. However this may
not be that useful since it is very unlikely that the fully linear
structure would ever be sampled.

As the RMSd values are really crude, you need to delve more deeply into
the analysis to understand what the differences in RMSd mean. This is
most easily done by looking at the structures and/or characterizing the
structures. You could look at energy, compactness, hydrogen-bonding, %
secondary structure, principal component analysis, and a host of other
properties.

The 2D RMSd plot is not so useful except to show that "close" structures
are sampled repeatedly. Clustering of the trajectory can be very useful;
this can be done with Moilview and soon with ptraj. Even without
application of a clustering algorithm, you can look at the 2D RMSd plot
and mark out regions denoted by the squares seen in the plot. Create
average structures over these local regions in time and then compare
visually to see the differences...

> maximum value as 2dRMS limit value. I have tried to use smaller limit
> and I got different results. That's the matter of limit value. Can

What do you mean smaller limit? I do not see any way in ptraj to specify
the limit; if you mean changing the maximal contour level, this shouldn't
change the results, just the visualization.

Good luck,