space filling mixture DoE design for cell culture media

[#1]

I am useing JMP 15.2 and I didn't see where to start space filling mixture design. Either from classical> mixture design or from custom design, I didn't see where to start a space filling mixture design

2. I got antibody yield data. I am wondering how may I put these data into JMP to do analysis. The data is as below

[#2]

The first thing I suggest that you do is to read the DOE documentation in the DOE Guide available in the JMP Documentation Library, under the Help Pull Down.  There is a section on Space Filling augmentation that should help you.

There is also a good number of Discussion items that can be reviewed.  If you search the Discussions for Space Filling, you will get a lot of previous discussions on the topic.

[#3]

In addition to what @txnelson shared, I'll address question #2. It looks like the image of the data you shared is in some sort of spreadsheet application, Excel perhaps? There are multiple workflows to get data from one application into JMP. I'll suggest the first thing I'd try since the data table is relatively small is a simple column header drag/copy, in the native application, of the entire spreadsheet, then open JMP and instatiate a new JMP data table. On a completely blank JMP table, just do a Edit/paste. Hopefully your JMP data table will populate and then I'd do some checking of the JMP data type/modeling type in the JMP Column properties menu to make sure each of the numerous characteristics are amenable to the analytics and data visualization workflows in JMP. If this simple copy/paste workflow doesn't work...there are other workflows that will. Just too many to write out in a discussion forum like this. My idea above for importation was just driven by 'how can I accomplish this task with the least number of key strokes?'

Welcome to JMP!

[#4]

Hello @loganshawn, and welcome to JMP !

In addition to the excellent responses above, I will add some interesting elements related to your questions or situation :

1) You can do Mixture designs both in the "Custom Design" and "Classical" -> "Mixture Designs" platforms, but you may not have the same options in these platforms. You mention a Space-Filling Mixture design, so this type of design seems to be generated from "Classical" -> "Mixture Designs" platform (see screenshot).

2) You can simply import an Excel file or PDF file containing table in JMP by clicking on "File" -> "Open" and then choose your file.
Since you will only import the raw data here (not the model or factor formating done previously in JMP), you will have to add some properties to your 4 factors (right-click on the column and then "Column properties"): Mixture, Design Role (Mixture) and Factor changes (Easy ?) for Space-Filling Mixture design. See screenshot.
Concerning the analysis/modeling, you can try to do it step-by-step to familiarise with the data:  visualize the results with "Graph Builder", try to see bivariate relations (one factor vs. yield) with "Fit Y by X" platform, and then try different modeling with "Fit Model". 

If you work with people who don't have JMP but still need to fill results/infos on a DoE (Excel) table, I would recommend to download and use the "Table Attributes Add-in". This add-in will help you export your DoE in Excel with informations about factor formating and model, so when you'll upload back the completed Excel file in JMP, you don't lose any information about how the model was created and how the factors were created. More infos here : JMP to Excel and back again. 

If you're not familiar with DoE, there are a lot of ressources to help you :
Statistical Thinking for Industrial Problem Solving | JMP
Design of Experiments Intro Kit | Getting Started with JMP
DoE documentation, and all the webinars and ressources on JMP community...

You can also try this JMP DoE Quizz that can help you figure out which type of design may be suited for your needs : DOE Quiz (jmp.com)

Hope this will help you,

[#5]

You have gotten several very good suggestions. As several of these messages point out, the amount of work involved will be more than normal since you do not have the original JMP file. JMP saves several things associated with each of the columns when creating a designed experiment. These items (called column properties) will make the analysis easier and work correctly. Moving the data to a spreadsheet removes those properties.

For example, I noticed that one of your first problems will be that with the spreadsheet, not all of the rows for your prototype percentages add to 1. This will be the first problem you will need to fix for JMP. In order for JMP to properly analyze a mixture experiment, the components must add exactly to 1. See rows 7, 8, and 9, for example (there are likely others). So check for rounding in your spreadsheet and make certain the four components add to 1. 

[#6]

This is how it would look in JMP with the appropriate column properties and corrections/manipulations mentioned by Dan Obermiller.

[#7]

Hi,louv, thank you so much for your effort. I really appreciate it. I am sorry that I have more questions.

1. Based on your attached file, when I clicked DoE dialog, there are some data table with random data. I am wondering if you have generated space filling DoE data table and pasted my data into that data table? 

2. I actually did the same way by generating DoE table through Classical>Mixture design> space filling. Then, I just simply copy and paste the date from my excel into the data table. However, the modeling result is different from the modeling result of yours.  From your model, when I maxmize desirability through prediction profiler, it shows that 0.85 X1 and 0.15 X3 will have maximum desirability. However, from my model, when I maxmize desirability through prediction profiler, it shows that 1 of X1 and 0 of X2, X3 and X4  will have maximum desirability. 

I am wondering why there are such difference from same data? 

BTW, what does "No intercept" mean in the below picture? I found that check or uncheck "No intercept" make big difference of modeling result

Thank you again. 

Best wishes

[#8 · Accepted solution]

The model which I have scripted includes two-way interactions. The data I entered was based upon your data table along with the column properties to reflect your setting ranges. The R2 for the model is improved with interactions. IMHO the best way to see if your model is useful is to try to verify via experiment.

 

[#9]

To add just a bit to the advice provided by @louv for verifying the utility of a model, in addition to additional empirical verification, I'd pause for a moment and examine the effects from a first principles or apriori knowledge point of view. Back in the day I'd ask the scientists and engineers I worked with something like, "If the model suggests that water runs uphill, I think the model is suspect and we may want to pause before entertaining it as some form of new learning or 'proof'." And especially in the apriori knowledge case, is the 'knowledge' factual and data derived or just urban manufacturing myth passed down through the ages because somebody saw something, years ago, but nobody can quite remember the exact circumstances or context? I've seen many a DOE study shatter these urban manufacturing myths. Good hunting.

[#10]

Hi, louv, thank you so much for your time. I learned a lot. 

I am sorry that I have a follow up question. I was trying to find lack of fit analysis and didn't see it. I am wondering if it is possible to do lack of fit analysis for mixture design. As I know there is lack of fit when there are replicates. 

Thanks