How to create a design that combines finding the best combination of substances together with finding their best concentration?

[#1]

Dear JMP team,

I am planning an experiment, where I have two categories of molecules (categoryA and categoryB). Both have two possible molecule candidates each: categoryA = {molA1, molA2} and categoryB = {molB1, molB2}. I would also like to model the concentration of each molecule as independent factor. 

I was first thinking to add two continuous factors like ConcentrationCategoryA and ConcentrationCategoryB. Here comes my first question. I have different high and low concentrations for the four molecules. Is it feasible to just use in the designs +1 and -1 as usual and just later on adapt the design table with the real (but different values), or do I somehow skew the design space with this.

Second I came up with the idea to add for each molecule one continuous concentration factor, resulting in 6 additional factors. In this case I get designs in which categoryA == molA2 and at the same time molA1 == +1. But since molA2 has been "chosen" it doesn't make sense to code for molA1 at all. Ideally it would have some kind of a "-" or NA. How can I build this and still have a solid design?

Due to time and device constraints I have only one run with max 48 experiments. Next steps will be validation in a larger scale set-up

Thanks a lot for your help,

Tom 

[#2]

Welcome to the community.  I do not understand your situation, but here are my thoughts/questions:

Are you interested in understanding causal structure our are you trying to "pick a winner"?  

I'm a bit confused, are the 2 Molecule Category factors independent?  Are both going to be part of the final product, or are you choosing between A & B?  Do you care about a possible interaction between A & B? Or are these part of a mixture (in which case you might want to use a mixture design)?

What model are you investigating?

One interpretation:  You have 3 factors, X1 is Molecule Category with 2 levels, A&B (Am I correct in assuming factor 1 is a categorical factor?). X2 is Candidate (can't tell if this is categorical or continuous), this may be nested? X3 is Concentration whose levels depend on levels of X1 and is continuous (this is definitely nested).  With nesting, you can determine the effects of each factor, but not interaction effects.

Y = X1 + X2[X1] + X3[X1, X2]

Another interpretation: You have 2 factors, X1 is molecule category and it has 4 levels.  X2 is continuous and X2 levels depend on X1 so it is nested.

Y = X1 + X2[X1]

Another, If molecule category is independent:

Y = X1 + X2 + X1X2 + X3 [X1, X3]

[#3]

Dear statman,

thanks for the fast reply. Let me further clarify.

My aim is to both pick the best combination of two molecules, each of which target a different biological pathway. At the same time I want to find out about relation between the pathways (synergistic, antagonistic effects) and ultimately also be able to tell which combination of 2 molecules and which concentration for the two is best. I would even like to propose a change in concentration of the winning tuple, by checking the coefficients of the concentration.

My model:
I have 4 factors. X1 is candidate from pathway 1 with 2 levels (A&B), X2 is candidate from pathway 2 with two levels (C&D), X3 is the concentration of  the candidate of pathway 1 whose levels depend on levels of X1 and X4 is concentration of candidate of pathway 2, the levels of which depend on X2. A,B,C, and D are given in total different concentrations, which creates my major point of concern. Could I just code X1 and X4 with (-1 and +1) and just manually insert the actual concentrations later, even if they are different conditionally by X1 and X2, respectively. 

In the meantime, I found this thread very related: https://community.jmp.com/t5/Discussions/Nested-DOE-with-continous-factors/td-p/674509 , just that I have this three times. What would you now recommend, because I didn't fully understand, how this thread was ultimately resolved. I also was not able to find out how to create a nested design with the custom designer.

Thanks,

Tom