Wanted: Non-linear Reg. w/Excel (update)

Ian Musgrave Ian.Musgrave at med.monash.edu.au
Sun Sep 29 20:08:45 EST 1996


G'Day All

Ian.Musgrave at med.monash.edu.au (Ian Musgrave) writes:
>From: Ian.Musgrave at med.monash.edu.au (Ian Musgrave)
>Subject: Re: Wanted: Non-linear Reg. w/Excel
>Date: Tue, 24 Sep 1996 03:32:59 GMT


>>I have a set of excel spread sheets to do non-linear regression 
>>for logistic (concentration response curves) and hyperbolic (saturation) 
>>curves. If you would like these examples, email me at the address in my 
>>signature, and I will email a ZIPed archive to you.

>Alternatively, you can download it from my homepage at

><http://www.mmcc.monash.edu.au/~ian-mu/nchem.htm>

>From the software section.

In the FIT.TXT "how to do it" file in FIT.ZIP, I mention inital values, 
precision and constraints, but do not elaborate. I have now fixed this. For 
those of you who have downloaded FIT.ZIP, here is the additional text.
=======================================
Inital values. If you choose an inital IC50/EC50 far 
from the true one, then the program 
may incorrectly fit the values (Find a local minimum, this is a 
problem for all curve fitters). Inspect the data, if the line does 
not follow the data, redo the fit with a new estimate of the 
IC50/EC50 (preferably one that is closer to the data than the 
"false" fitted value").  

Precision is important. If for example, your concentration values range 
from 1e-4 to 1e-9 with an IC50/EC50 around 1e-7, then you should 
adjust you precision (solver | options) to 1e-9. The precision should 
allways be a factor of 100 greater than the expected IC50/EC50. If 
your IC50/EC50 does not change when you do a fit, suspect that the 
precision is too low.

Constraints are also important. Because of the nature of the fiting 
process, the program may sometimes try absurd values, and get stuck. 
So you need to set lower (and sometimes upper limits) for the program 
to use. For concentration response curves, lower limits of the IC50/EC50 
values should be around 100 fold lower than the lowest value on your 
curve (in the example above a constraint of IC50 >= 1e-12 is 
appropriate). The constraint for slope should be around 5. The 
constraints are set in the constraint dialog box of the solver 
utility. Note that there seems to be a bug in the macintosh 
implementation of solver, in that you can set a constraint, but 
cannot edit or delete it ( I can't tell if this is due to a resource 
hungry excel running on our old mac, or a true bug).
============================================
Cheers! Ian
---------------------------------------------
Ian Musgrave Ph.D, Prince Henry's Institute of Medical Research
PO Box 5152, Clayton 3168, Australia.
Phone +61 3 550 4286 FAX +61 3 550 6125
Lab: Ian.Musgrave at med.monash.edu.au <http://www.mmcc.monash.edu.au/~ian-mu/nchem.htm>
Private: Reynella at werple.mira.net.au <http://werple.mira.net.au/~reynella/>




More information about the Bio-soft mailing list