From: John Philo <jphilo@mailway.com>
  To  : christopher chin <chinchiquei@yahoo.com>
  Date: Thu, 6 May 1999 14:11:00 -0700

RE: Compare fitting program for Ka

Chris,

It is impossible to judge whether any of these differences between
programs are significant because you haven't included any error bars
for the parameters. If the Ka numbers from different programs differ
by 3%, but the uncertainty in that number is 30%, then in my view who
cares?

The Beckman software won't give you error bars for a non-weighted fit
(for no good reason, especially since every fit of interference data
with their software is really a non-weighted fit) but WinNonLin
certainly will.

You are posing a reasonable and practical question about how to judge
whether a mutation makes a significant difference, but again that
question can't be discussed without knowing the uncertainties in the
values.

You also need to keep in mind that a difference of 30% represents a
very small difference in binding energy (less than thermal energy).
Often Ka's reported in the literature by various methods have
uncertainties of 2-fold or more (and data of that quality has answered
many important questions in biochemistry!) Thus suppose one could
measure the Ka of mutant and wild-type with a precision of 10%, and
the numbers differed by 30%.  That difference would be STATISTICALLY
significant, but I think most people would agree that it would
represent a fairly trivial difference in binding thermodynamics and
would NOT indicate that the residue that was mutated plays a
"significant" role in the interaction.

The differences you are seeing with regard to picking the correct
assembly stoichiometry are another matter and might indicate a more
serious problem. But again one really can't judge the situation based
on variances alone. It may well be that ALL of these fits are bad in
the sense that the residuals are systematic. If neither the
monomer-trimer nor monomer-tetramer model is sufficient to explain the
data, then I wouldn't be surprised that different programs might pick
a different (but equally wrong) model. It also may be true that the
experiments simply do not cover the proper range of concentrations
with sufficient signal/noise for any program to be able to truly
discriminate between monomer-trimer and monomer-tetramer.

In my view a better approach to testing the programs is to feed them
simulated data to which you have added random noise, because then you
know the 'correct' answer. With real data (whether yours or Beckman's)
one never knows whether the results are influence by sample
heterogeneity or systematic errors.

John Philo

-----Original Message-----
From: christopher chin [chinchiquei@yahoo.com]">mailto:chinchiquei@yahoo.com]
Sent: Thursday, May 06, 1999 11:35 AM
To: RASMBer BBRI
Cc: James Lee
Subject: Compare fitting program for Ka


Dear RASMBers,

Several months ago when we upgraded our Beckman Origin
software from version 3.78 to 4.1 (4.01?) I was
concerned with the inconsistency of the two versions.
It turns out that the inconsistency was due to the
fact that I was probably unknowingly using the wrong
files for comparison, thanks to John Philo's
suggestion.  Upon recalculating the data, I found that
the two versions of Beckman produce results that are
very close  (about 3% or less).  I think I can rest
easy with that kind of number.  At that time Tom Laue
also suggested that I use Nonlin.  I then tried to
refamiliarize myself with this software which I
learned to use at the U Conn training course.
However, when I compare the WinNonlin result with that
of the Beckman 4.1 program, using one speed, three
concentrations to do a global fit, the difference
between Beckman and WinNonlin is about 2.7%, I have no
problem with that.  However when I used 3 speeds, 3
concentrations for global fit (total of 9 files), the
difference is close to 30% (2.33 vs. 1.62, comparing
apples to apples applying non-weighted fit function
for both.  Can anyone tell me how to use weighted fit
for WinNonlin?).  This bothers me.  Can someone point
out where I made a mistake? If indeed, I did make a
mistake.  I would like also to thank Marc Lewis for
the suggestion that I use Matlab or MLAB.  I am not
familiar with either of these programs.  So before I
invest my time and resources searching for another
"perfect fit" program, I would like to ask those of
you who are familiar with these programs to fill in
the blanks using the 9 data files that were provided
in the attachment.  This would be an exercise, if you
were, to see how much variation in Ka one can tolerate
(we need to decide on a gold standard--can we use
Beckman's Origin software as the standard, since they
manufacture the instrument, and to see how other
software programs pan out?)

I am curious as to what would happen if one tries to
use the UC method to study association behavior
between WT-protein and Mutant-protein--if the
variation in Ka between the software used from lab to
lab is greater than the small difference between the
samples under study.  Can the same conclusion be drawn
from different labs using different software programs?

 	Also in one SE experiment when I tried to model the
association system to see if the system is going from
monomer to dimer, monomer to trimer or monomer to
tetramer. I used "Goodness of fit " or "Variance" as
an indicator from Beckman's Origin 4.1 and compare
that with "Square root of variance" as an indicator
for WinNonlin.  I found that with the Beckman's Origin
4.1 software, the system favors monomer to tetramer
(for both, weighted fit as well as nonweighted fit),
while with the WinNonlin software, the system favors
monomer to trimer (I do not know how to do weighted
fit using WinNolin).  Could this be an effect due to
variation in different software used? (See summarized
result in attachment 3; I also include the raw data
files for those of you who are interested in
recalculate the data).

Regards

Chris

-------------------------------------------------------
Christopher Chin
Manager
Sealy Center for Structural Biology
HBC&G, 5136 MRB. rt1055 UTMB
cchin@utmb.edu, 409-772-1693, efax 708-585-1920
-------------------------------------------------------




===
-------------------------------------
Christopher Chin
Manager
Sealy Center for Structural Biology
HBC&G, 5136 MRB. rt1055 UTMB
cchin@utmb.edu, 409-772-1693, efax 708-585-1920
-------------------------------------
_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com