Formulation optimisation, custom design, how to choose the right high levels

Question

Hello lovely jmp community, I’m a very new learner and was hoping I could pick some brains for insight or even get a link to a chat somewhere on the forum I may have missed in my search!

Soooo…. I’m planning to use a custom design in jmp to screen 10 continuous factors (excipients) for formulation optimistation. The plan going forward once our significant linear, quadratic and/or interactive effects have been identified is that we will then go on to RSM. Historically, univariate optimisation has been used in my company and as such they have had the capability to investigate factors at fairly high levels (concentrations of excipients). I’m unsure how the high levels of factors will work in combination though and ultimately, we may have situation where our active “crashes” out of solution which would then render any response analysis unachievable to then feedback into the design. So now I am in two minds of whether I

* Make a design where the factor levels are approx. half the concentration of what they typically investigate univariately in the hope I can identify X1, X2, X1X2 for further optimisation and raise the factors higher level at the RSM stage OR
* Go with the higher levels from the get go, potentially risking the loss of response data at these high level combinations and if it is a case where we don’t get read outs then try reducing the higher levels at that stage and rescreen?

If it helps the design I’ve generated so far has 74 runs total (of which 6 are centerpoints) and I can go upto a max of 96 runs in one go. As I mentioned I’m quiet a new learner to jmp so if there is any advice/reading ,material you could help me with it would be very much appreciated!

Byron_JMP · Answer

You mentioned a couple of details about your formulation experiment that might be interesting to explore a little further:

I was wondering if you were also thinking about mixtures?

Are you thinking about order of addition experiments?

Mixture designs might help to define an area of your response where your API crashes.

gg1987 · Answer

Hi Byron,

Thank you so much for the response, I was actually toying with the thought of a DSD but hadn't thought of being able to do the reps on the one plate instead! I'll give that a bash and thank you very much for the insight :)

Byron_JMP · Answer

I'd like to suggest a little different approach to the first screening experiment.  In JMP there is a Definitive Screening Design (DSD) that might save you some time and effort. If your 10 factors are all continuous, then the DSD design is 25 runs rather than 74. Even adding 8 extra runs and 2 blocks with center points only costs 30 runs. (If you have a robot to do the pipetting, you could do 3 replicates in one plate. Replicates that are averaged together in the analysis step are nice for reducing the variation due to biologics positional effects on plates)

From this first experiment you should be able to get an idea about the relative importance of your 10 factors, so that in your next RSM design you can work with a subset, and better ranges.

Ranges.. In screening designs, "go big or go home" is my motto. In the first screening design and to a degree, in the first RSM design, I want to identify the edge of failure so that I can nail down (with data) a proven and acceptable range.  Once I have that, then another RSM, maybe with even less factors (things that are critical to quality) and ideally an I-optimal design that focuses on the middle of the design space for finding optimal operating conditions, might bar useful.

just my $.02

-B