<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=us-ascii">
<TITLE>Message</TITLE>
<META content="MSHTML 6.00.2900.2802" name=GENERATOR></HEAD>
<BODY style="MARGIN: 4px 4px 1px; FONT: 10pt Tahoma">
<DIV><SPAN class=147182916-27012006>Just to set the record straight, Jack has
apparently misinterpreted my paper (Anal. Biochem. 279, 151 (2000). It does not
say that averages higher than Sw are too noisy to be useful, nor
did it attempt to demonstrate that by simulations. First, the work of
Jack and others clearly shows that to be false. Second, if that were my
opinion, why would I have bothered to develop a new algorithm that gives better
estimates of the errors in Sw, Sz, etc. or to include those
computations in my software? </SPAN></DIV>
<DIV><SPAN class=147182916-27012006></SPAN> </DIV>
<DIV><SPAN class=147182916-27012006>What the paper points out is that when using
the standard way of calculating the error bars for the g(s*) distribution (the
original Stafford algorithm), as you include more scans in the computations the
error bars do not reduce as rapidly as one would expect from fundamental
signal/noise considerations (i.e. the standard algorithm overestimates the error
bars and thus underestimates the true precision of Sw, Sz, etc.). The revised
algorithm described in that paper, and optionally implemented in
DCDT+, partially corrects this problem.</SPAN></DIV>
<DIV><SPAN class=147182916-27012006></SPAN> </DIV>
<DIV><SPAN class=147182916-27012006>Also I should clarify that I was wrong to
imply in my earlier response to Holger that one cannot calculate a formal Sn
from g(s*), c(s), etc. My program DCDT+ does not calculate that one simply
because no one has ever asked for it, but in response to this discussion I will
implement that one too in the next release.</SPAN></DIV>
<DIV><SPAN class=147182916-27012006></SPAN> </DIV>
<DIV><SPAN class=147182916-27012006>John</SPAN></DIV>
<BLOCKQUOTE dir=ltr style="MARGIN-RIGHT: 0px">
<DIV></DIV>
<DIV class=OutlookMessageHeader lang=en-us dir=ltr align=left>-----Original
Message-----<BR><B>From:</B> rasmb-bounces@rasmb.bbri.org
[mailto:rasmb-bounces@rasmb.bbri.org] <B>On Behalf Of </B>John
Correia<BR><B>Sent:</B> Thursday, January 26, 2006 12:29 PM<BR><B>To:</B>
Walter Stafford; Holger Strauss; rasmb@rasmb.bbri.org<BR><B>Subject:</B> Re:
[RASMB] averages from SV<BR><BR></DIV>
<DIV>Holger</DIV>
<DIV> </DIV>
<DIV>To my knowledge my review from a few years ago is the most
recent comprehensive overview of weight average. It finishes
with a mention of other averages & who has applied them. </DIV>
<DIV> </DIV>
<DIV>J.J. CORREIA (2000) ”The Analysis of Weight Average Sedimentation Data.”
Methods in Enzymol., vol 321, 81-100.</DIV>
<DIV><BR> </DIV>
<DIV>Soon after I published that review John Philo published a paper
claiming typical data is too noisy for higher averages, and demonstrated
so by simulations. As Walter mentioned these averages are also
calculated in the DCDT portion of SEDANAL. Peter Schuck also suggests
they are too noisy. Attached please find a test of those claims, derived
from XLA data using g(s) analysis in SEDANAL. Clearly if done properly
they are not too noisy! (I have compared this analysis with results from
Philo's DCDTplus2 and the error bars are comparable for Sw, Sz and Sz+1 - John
doesn't report Sn.) </DIV>
<DIV> </DIV>
<DIV>As to claims about no theoretical meaning, moments or averages of
distributions are what they are. One could easily write a program
to fit these moments and compare them to some model. We have done this
but I generally only attempt such complexity when the system is very
complex. I find it much more useful to compare weight average fitting
with direct boundary fitting. For my self association systems the K's
derived typically agree to within 20%. The direct boundary fitting
usually reveals what causes the disagreement, typically aggregation at the
centrifugal edge of the boundary.</DIV>
<DIV> </DIV>
<DIV>Applying these averages to different signals has its pitfalls,
but Schachman 1st described absorbance average moments in his book, so
the use of different signals is an old idea and easy to apply as a
fitter. We again prefer to resort to direct boundary fitting,
combining different wavelengths and extinction coefficients to different
components and species. I strongly agree the information is in the shape
of the sedimenting boundary. (This does not necessarily mean the shape
of the derived distribution.) Comparisons to averages should in
principle work with the usual caveats about reversibility, degree of
aggregation, the presence of plateaus, etc. </DIV>
<DIV> </DIV>
<DIV>I hope you try this approach, apply your own direct statistical
tests, and convince yourself of its potential utility. I have no doubts
there may be a system dependency. My bias continues to be
direct boundary fitting is the most informative, although we use weight
average for model building, extrapolation to end points, and estimation of K's
for initial guesses.</DIV>
<DIV> </DIV>
<DIV>Good luck!</DIV>
<DIV> </DIV>
<DIV> </DIV>
<DIV>-------------------------------------------------------------------<BR> Dr.
John J. "Jack" Correia<BR> Department of Biochemistry<BR> University
of Mississippi Medical Center<BR> 2500 North State
Street<BR> Jackson, MS 39216<BR> (601)
984-1522
<BR> fax (601)
984-1501
<BR> email address: <A
href="mailto:jcorreia@biochem.umsmed.edu">jcorreia@biochem.umsmed.edu</A>
<BR> homepage location: <A
href="http://biochemistry.umc.edu/correia.html">http://biochemistry.umc.edu/correia.html</A><BR> dept
homepage location: <A
href="http://biochemistry.umc.edu/">http://biochemistry.umc.edu/</A><BR>-------------------------------------------------------------------<BR> <BR> <BR></DIV></BLOCKQUOTE></BODY></HTML>