Design space regions do not match.

[#1]

To practice using JMP, I am analyzing and reproducing data from papers.

Among these, screening using the Plackett-Burman design and optimization using the central composite design were performed for three impurities to create a predictive model.

Similar predictive model equations were created for all of the models described in the paper.

The paper states that pooling was performed during the optimization process, but a model including all parameters was used to create a design space that combined the models.

Following this, we created a predictive model equation without pooling, i.e., including all parameters, to create a design space.

However, the area of ​​the design space where impurities are 0.1% or greater does not match, and my design is wider.

The equation for the predictive model used to create the design space is not disclosed, so it is unclear, so what is the reason for this?

Sorry for the poor writing.

[#2]

Hi @keita_,

It's a good idea to practice with paper providing data. However, the modeling strategy and choices are often not detailed enough to reproduce the results, and even if the level of information is sufficient, you can expect some deviations from the paper, see a similar post here Replicating DOE analysis from a paper where most of the results are similar, but not the validation.

It's very hard to help you without more in-depth explanations (for example more context about the graphs displayed, the contour zones, etc...), a dataset to show what you already have done, and more details about the modeling and choices done on the mentioned papers (maybe linking the paper or citing the reference could also help).

I would think that in your case, a difference in the model fitting/equation (maybe because of the difference in "pooling", not sure what does this mean in this context ?) can explain the differences you observe, since the equation for the predictive model used to create the design space is not disclosed in the paper.

Hope this discussion starter may help you,

[#3]

Dear. @Victor_G 

Thank you for replying.

Details are given below.

--

The papers referred to are listed below. Not open access. Sorry.

Application of a Statistical Approach to the Process Development of Futibatinib by Employing Quality...

The following is an extract from the section on design space creation.

’’Since the CPPs and two-factor interactions affecting 5, 6, and 7 in futibatinib were clarified, the DS was constructed based on these multiple regression equations. Since all parameters were included in the three multiple regression equations obtained by these analyses, the DS was constructed from the prediction model by including all parameters in the multiple regression equation (Figure 27). Based on this result, it is clear that 5, 6, and 7 in futibatinib can be controlled sufficiently by target values and ranges of the process parameters. However, 7 was proved to deviate from the acceptance criterion when manufacturing is conducted by the combination of internal temp_3 at 35 °C and stirring time_2 at 5 h. The parameter combination of the worst case for generating 7 was estimated, and the amount of 7 in futibatinib was predicted under the condition (Figure 28).’’

Am I correct in assuming that this means that the common factors (main effects, interactions and quadratic terms) are included in the design space creation, each of which saves the predictive model equations, and from which the contour profiles are created?

On a separate note, I'd like to add a color to an arbitrary area on the design space on JMP.

I would be grateful if you could let me know.

Sincerely yours.