cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Register for our Discovery Summit 2024 conference, Oct. 21-24, where you’ll learn, connect, and be inspired.
Choose Language Hide Translation Bar
joepark
Level III

How to do Fit Y by X if I want to do for factorial Xs

Hello,

I am trying to analyze cell survival in response to multiple drugs to understand synergistic effects

For example:

X1

X2

X3

Y

100

100

10

5

50

50

5

25

200

200

0

10

100

100

0

30

200

0

10

3

100

0

5

20

0

200

10

1

0

100

5

8

0

0

0

60

 

Is there way I can do Fit Y by three different drugs X1, X2, and X3 on JMP?

Initially I ran Fit Model and did prediction profilers, but the prediction values are very off from the observed values.

I applied the Logit/LogitPct/and all other transforms to the Y response in the Fit Model launch dialog but the prediction values are not close to the observed values.

Is there way I can analyze Y as response to factorial Xs?

 

5 REPLIES 5
statman
Super User

Re: How to do Fit Y by X if I want to do for factorial Xs

Joe,

It is helpful if you just add the JMP file.

I created the data table in JMP and did a fit model (started with a saturated model and then reduced it).  I saved the prediction formula to the data table (attached)

 

I have no idea what you are experimenting on, whether your data is of any practical value or whether the results make any scientific sense so difficult to provide advice.

"All models are wrong, some are useful" G.E.P. Box
joepark
Level III

Re: How to do Fit Y by X if I want to do for factorial Xs

Hello statman,

Thank you for the reply. Unfortunately, I am not allow to share any data without permission.

My apologies for the lack of supplements.

 

I am analyzing synergistic effects of multiple drugs.

My first approach was Fit model then had the prediction profilers. Then I found that the prediction values are very off from the experimental values.

Yesterday I asked questions about this and got some great feedbacks (transforming responses to Logit in Fit model dialog)

After transforming responses, the prediction profilers are more close to the experimental values but it is not that close enough.

Here is my original post for the reference (https://community.jmp.com/t5/Discussions/Prediction-profiler-prediction-values-do-not-match-with-the...)

 

I am wondering if I can do Fit Y by X. In my case, I have 3 factors (Xs) and one response (Y). Each X has different range of concentration. 

So like full factorial I want to do

Fit Y by X1*X2

             X1*X3

             X2*X3

             X1*X2*X3

Is it possible to do on JMP?

 

Re: How to do Fit Y by X if I want to do for factorial Xs

I am not sure why you cannot use Fit Least Squares. I also captured your example like @statman and analyzed your data this way. I get a reasonable model. The X1 factor appears to have no effect. The X2 and X3 both have effects and an interaction, which is what you wanted to know. Here is the actual by predicted plot, which exhibits reasonably good agreement between the observations you posted and the reduced model that I used.

 

actual by pred.PNG

Re: How to do Fit Y by X if I want to do for factorial Xs

Another idea is to use a probit analysis with JMP. You measure cell survival, so if you know the number of cells at the start and the number still alive at the end, you can use the linear predictor with these counts.

 

I do not have an example like your data, but I can illustrate with the Ingots2 data table in the JMP Sample Data folder. Here is the data table:

 

table.PNG

 

We have the count of those ingots that are ready, not ready, and the total. This is analogous to your cells that are alive, dead, and total. Next we open the Fit Model dialog and make these changes:

 

dialog.PNG

 

The initial model includes both factors and their interaction. Your factors would be the drugs instead. The results for this model show that the interaction here is not significant.

 

full model.PNG

 

The insignificant terms are removed and the reduced model (in this case) has only one factor.

 

reduce.PNG

 

We can see that as the factor level increases, the proportion not ready also increases. You can also use a profiler with this model.

SDF1
Super User

Re: How to do Fit Y by X if I want to do for factorial Xs

Hi @joepark ,

 

  Using your example data table, you can do all kinds of fit Y by models. You can do them individually using the Fit Y by X platform and then do a linear (or non-linear fit) to each of the Y by X1, Y by X2, and Y by X3. You can switch back and forth real easily by using the column switcher. Go to the red hot-button next to the bivariate fit > Redo > Column Switcher. Then select X1 in the upper panel and all the other X's in the lower panel, then hit OK. This should answer your first question, but will not answer any synergistic effects.

 

  I'm not sure I really follow your question on how to analyze Y as response to factorial Xs. Was your DOE originally based on a factorial design? If you're trying to do a Fit Model and have the model effects be defined by a full factorial, after launching the platform, and selecting X1 to X3, click on the Macros option and select full factorial. Then you can pare down the model effects to get a reasonable fit. Take note of the effects that have the "^" carat next to them. This means that high order terms (those containing things like X1*X1 or X1*X2 would be appearing before the main order effect like X1. Using your data table, it suggests that X1 is not a significant factor and wants to remove it from the fit. You would need to assess whether this makes sense or not.

 

  You might need to expand your DOE or run replicates in order to reduce the noise in your system. The prediction profiler is helpful, but what's more important is whether your fit model and data always lie within the +/-95% confidence interval. You can get to this from the red hot-button and save those as columns. In the test case I did, all data (Y and prediction values) fall within these CIs. You might need to rethink your model to make sure that it's accurate enough.

 

Hope this helps,

DS