fitting with dynafit?

Dima Klenchin klenchin at REMOVE_TO_REPLY.facstaff.wisc.edu
Fri Aug 10 08:52:03 EST 2001


Jim Reid <j.reid at sheffield.ac.uk> wrote:
>Hi,
>
>I've cc'd this to bionet.matabolic-reg as it seems to be the closest to
>a "numerical modeling in enzymology" newsgroup.  
>
>Anyway I suppose it very much depends on what you are trying to do. 
>I've fitted binding data without too many problems but that isn't very
>tricky.  When I've tried to fit more demanding data sets (e.g. progress
>curves at varied ligand and enzyme concs.) I've had more problems.  On
>the basis of _very_ little experience (you have been warned!) these can
>be due to problems analogous to determining Vmax when [S] << Km. 

No matter what fitting and how it is performed, this simply should never be
done because any number obtained will always be meaningless. 

>As for the initial value problem, I have suffered from this when trying
>to simulate the first few minutes of the time course of an enzyme with a
>long lag phase in an attempt to get a steady state rate.  

The original question stated that the problem is limited to Dynafit, while
Sigma Plot and Excel, using the same guesses, converge easily. 

I have no idea. Either Dynafit uses a different least squares algorythm 
(Sigma Plot uses Marquardt-Levenberg which is, I think, a "standard")
or it is a bug in Dynafit. 

>convinced that this is anything to do with the program (similar problems
>with kin/fitsim), but that there are a lot of local minima and picking
>good starting values is just essential.

Actually, if the data are reasonably good and the function is chosen 
correctly, initial guesses practically do not matter. All such problems 
occur mainly when people try to do impossible things.  (Like, I've seen
a guy insisting on fitting five experimental points to a function with 
four parameters. That "result" is even published in JBC :<) 

>Are they worth the effort?

Non-linear regression? A god send - considering alternatives of linear
transformations. 

>Always? Sometimes - It depends? Never, just design expts. to avoid
>them?  

Like with any tool, it works where it is designed to work. 

        - Dima








More information about the Methods mailing list