Fit Multiple non linear Y responses

[#1]

Hi all, 

I want to do a Fit model with a least squares regression and I want to include the effect of the dilution of antibody which follows more or less a sigmoid model. I could of course include dilution up to to dilution^4 as effects, but a polynomial model doesn’t really give the output I want since I would like the response plateau from a sigmoid model to be present. Essentially, I want to do the same as a non-linear fit using the fit non-linear platform, but include other terms such as interaction terms. Would this be possible in the context of the least-squares model and if so, how would I go about doing this? 

Best regards,

Frederik

[#2]

You should see Analyze > Specialized Modeling > Fit Curve for a platform that is better suited to the logistic curves. See Help > Books > Predictive and Specialized Modeling for a chapter about Fit Curve. It is loaded with help and examples.

[#3]

My apologies! In my haste, I failed to recognize that your model includes more than one predictor. Yes, you must use a custom model with the Nonlinear platform for that purpose. Use the same book as I mentioned before but see the chapter about Nonlinear.

Someone else might have an idea about how to use Fit Model to begin with in this case, but I can't think of one at the moment.

[#4]

Without completly understanding your experiment....

I might try to fit a non-linear curve (like a 4p or 5p) to each of the samples and the take the parameter for the upper asymptope or the inflection point (ec50) or whatever is relevant and use that as a response in the linear model with the other imputs that were used to effect the dilution curve.  

For example if I was looking at coating, block and wash buffers, and a couple of dilution ranges. I could fit each dilution, get the inflection point and upper plateau from the non-linear fit. and then go to Fit Model and include the four parameters and their interactions as model effects and use the two parameters from the non-linear fits for the responses.

[#5 · Accepted solution]

Frederik,

Have you thought about trying Neural Nets?  The Tan H function is sigmoidal in nature and you can use as many inputs as you like similar to Fit Model.  I would suggest doing a series of fits with different numbers of hidden nodes and then go with the model that is the least complicated, but still meets your prediction/fit criteria.

HTH

[#6]

Hi Bill,

Thanks for your suggestion. I have used TanH based neural network models before, but had not considered it in this context. I will try it out. 

Thanks to the other repliers as well!

Frederik