[RASMB] HELP!

[Arthur Rowe]

Hi Chris

Looking at your graph, there is enough indication of where the peaks are for
fitting to be stable and possible. Given, of course, software that is
appropriate.

The problems are

(1) there seem to be 8 peaks present, which means that you have to float 8 x
3 = 24 parameters (assuming that you fix the baseline - see below). {3
parameters, area, location along x axis and SD, per peak}

(2) most curve-fitting packages have a multi-gaussian fitting option - but
use with care! They may use unfriendly* parameters such as 'peak height' and
what they mean by SD can sometimes turn out to be the variance!

(3) even having got the option that you want for fitting, entering some
vague guesses for the parameters and then calling up Marquadt-Levenberg for
fitting (and major packages such as ORIGIN give you no choice) then you will
probably get nothing but error messages. Agreed that lots of times fitting
in that way poses no problems, but with multi-gaussian fits it is unstable

(4) the baseline - is it really level, or does it go up? Obviously I am not
saying that the elevation at the RHS is not real, very possibly it is. But I
find that if a baseline gradient is present for real, then stripping it off
gives a much better and more valid fit.

*OK - these parameters may be very friendly for other purposes

Suggestions:

Option A. You have access to an Apple Mac

Download Pro Fit from http://www.quansoft.com/ - you will have to pay for
the key, but this software at $95 is many times lower in price than ORIGIN.
And it is better in lots of ways - e.g. you have a choice of three
alternative fitting algorithms selectable, a lovely simple programming
language if you want to use it (not just C !), interface to any compiled
language allowed, and so on . . . .

I still do not like the way their multi-guassian fitter module is set up,
but I can mail you our own locally written one. If you would like a demo of
what it does - just mail me the textfile of your data. The way I use it,
ROBUST fitting (see - if need be -
http://www.clecom.co.uk/science/autosignal/help/Robust_Fitting.htm for what
I am talking about) is indeed ROBUST, and gets me very near the answer, and
yields excellent first guesses for final Levenberg-Marquadt fitting. NO
crashes.

Option B. You have only WINDOWS software available

Leaving aside the obvious point that you can buy an inexpensive Mac plus Pro
Fit software for around the price of 1 copy of ORIGIN, there is a decent
alternative available in the cross-platform Sigma Plot ($495 academic rate -
still expensive!): (http://www.rockware.com/catalog/pages/sigmaplot.html). I
cannot, I am afraid, remember what their multi-gauss fitter is like, but
assuming it is set up with the parameters you want, the general approach is
as follows:

(i) make the very best guesses you can for the position of the center of
each peak
(ii) assuming this is an input parameter, FIX it for each gaussian, but
FLOAT the areas & SDs
(iii) using the estimates for the latter as first guesses, float everything
and pray!
(iv) if all that fails, then back to (i) with better initial guesses

Only general problem with Sigma Plot - otherwise a nice if overpriced
utility - is that plotting user-defined functions is klunky beyond belief.

Hope all this helps. And mail me your data if you like.

All best

Arthur

--
*******************************************************
Arthur J Rowe
Professor of Biomolecular Technology
NCMH Business Centre
University of Nottingham
School of Biosciences
Sutton Bonington
Leicestershire LE12 5RD   UK

Tel:        +44 (0)115 951 6156
            +44 (0)116 271 4502
Fax:        +44 (0)115 951 6157
email:      arthur.rowe at nottingham.ac.uk
            arthur.rowe at connectfree.co.uk (home)
Web:        www.nottingham.ac.uk/ncmh/business
*******************************************************




Dear RASMB colleagues,
 
Does anyone know of a software program that can deconvolute mutiple peaks
that are not well separated? (Please see the attachment)
Your help in above regard is greatly appreciated.
Many thanks in advance.
 
Regards,
 
Chris
 


--------------------------------------------------------------------------
Christopher Chin 
Manager, XLA-Analytical Ultracentrifugation facility
Sealy Center for Structural Biology
HBC&G, 5.134 MRB.UTMB, Galveston,Tx 77555-1055
cchin at utmb.edu, 409-772-1693, efax 708-585-1920
---------------------------------------------------------------------------




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.rasmb.org/pipermail/rasmb-rasmb.org/attachments/20030904/833b9e42/attachment.htm>