fitting with dynafit?

Pedro Mendes mendes at vt.edu
Fri Aug 10 10:13:27 EST 2001


At the risk of making some publicity to my own work:

Least squares problems, especially with many data points and many 
variables, have many local minima in the error surface. This makes
solving these problems with Levenberg-Marquardt (or other gradient
descent methods) a challange. You need to be really close to the
solution for these methods to converge. With the program GEPASI
(which I am the author of), you can do the fitting with many different
algorithms, including Levenberg-Marquardt . Importantly Gepasi contains
Genetic Algoriths, Evolutionary programming and Multistart (with L-M) 
which all have a much bigger chance of solving your fitting problem.
The only drawback to these methods is that they will take longer time
to run.

Gepasi is available for free at http://www.gepasi.org
There is also a paper in the journal Bioinformatics discussing these issues 
(and optimisation in general):

Mendes & Kell (1998) "Non-linear optimization of biochemical pathways: 
applications to metabolic engineering and parameter estimation." 
Bioinformatics, 4(10):869-83

Abstract in PubMed:
http://www.ncbi.nlm.nih.gov:80/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=9927716&dopt=Abstract

Disclaimer: I am the author of both the software and the paper that I 
mentioned above, so my opinion is definitly biased...

-- 
Pedro Mendes
Assistant Research Professor
Virginia Bioinformatics Institute
Virginia Tech
http:/www.vbi.vt.edu/~mendes




More information about the Methods mailing list