Design of experiments

[#1]

Hi everyone, 

first of all, excuse my english because I am not a native english speaker. I am a beginner in the field of design of experimens and this is why I am looking for insights. My problem is the folowing: 

1. I need to optimize the concentration of 7 proteins contained in the cell culture medium we use

2. We know that all these 7 proteins have an significant effect on the response ( yield)

3. We don't know if there is any interaction between the proteins 

4. We know that the effect of some proteins is not linear to the concentration: this means that higher concentration than usual can have a negative impact on the yield ( possible quadratic effect ) 

5.  We also know that the cells used as starting material for cell culture have a big impact  on the yield, but since we can't control that factor, we are thinking about using it as a blocking factor. 

My questions are the following : 

- How do we start ? do we evaluate the blocking factor in the first plan  we make ? 

- Do we need to use three level per factor ( concentration of protein) or two to reduce the number of assay ( but we could miss quadratic effect) 

Thank you in advance. 

[#2]

Hi @Ekokase,

We probably can't cover all the things that you should think about before designing and executing your experiment. But I am happy to provide some ideas.

Some important questions:

How many assays are you able to run? 

How many cell starting material batches do you have?

How many assays can you run from each batch of cell starting material?

A Definitive Screening Design might be a good starting point. This would provide a relatively small experiment (e.g. the default design from JMP for 7 factors with 3 blocks and 3 centre points is 23 runs).

Having said this, you already know that all factors are important, so this is not really a screening situation. Instead you might want to design a Custom Design for the RSM model.

I hope this helps,

Phil  

[#3]

Welcome to the community.

Adding to Phil's questions:

1. Have you studied the measurement system?

2. How do you KNOW the 7 proteins have a significant effect?  How were they studied and under what inference space (what was the noise (e.g., starting material) doing when they were studied?)

3.  How do you know it is non-linear vs. this is an hypothesis you have and want to study?

4. If you know starting material has a huge effect, how do you want to handle this in the future? For example, do you want you become robust to it?  Do you want to adjust your process to compensate?  Do you want to work on improving the starting material?

As Phil suggests, there are many things to contemplate and consider before selecting the appropriate design.  AND the likelihood of you choosing the exactly correct design is near 0 (even for very experienced designers).  "the best design you'll ever design is the one you design after you run the experiment".  So think sequential.  The first experiment will provide information to design a better experiment.  At any point in your iterations, create multiple designs (easy to do in JMP).  Evaluate what knowledge each design may provide vs. the cost to run it.  Anticipate every possible outcome for each and then select one, get data, analyze and iterate.

[#4]

Thank you for taking the time to answer my question. 

A little bit of context: We use stem cells (our starting material) and differentiate (transform) them into adipocytes and theses 7 proteins (cytokines) are necessary for that differentiation. Our response is the percentage of adipocyte. 

1. Yes, the measurement system is validated.

2. The choice of these proteins is justified by years of research and these proteins are known, in the bibliography, for having an impact on this differentiation. But you have a point, I don't think that we can be 100% certain that all are important since this kind of study was not done before (assessing the effect of each factor). 

3.  Some cytokines are reputed to not have a linear response according to their concentration. For instance, the increase of the concentration may have a negative effect on response. This is the case for one of our cytokines.

4.  Yes, the finality is to have a combination of cytokines that give satisfying results though the starting material, but I don't think we could adjust the process since it is unknown what make some stem cells have a better yield than others. Our hope it to have at least 55% of adipocyte in all cases. 

[#5]

That is all useful to know. Thanks, @Ekokase. Are you able to tell us about the feasible number of assays, cell material batches, and assays per batch? This will be important.

[#6]

Yes I can  @Phil_Kay , 

18 assays per design is the maximum that can be done technically if it has to be carried out on the same day. We have enough stem cells yet doing 70 assays using only one cells batch would not be possible, the average number of assays that can be done is around 30 (  the cells batches do not have the same size in term of cells quantity).

[#7]

In which case, you will probably want to use Custom Design with an RSM model for the protein factors and a blocking factor to manage cell material batch. It really depends how many runs you want to do in total, how many cell material bxs you want to look at, how many runs you want to do per day, if you can look at multiple cell material bxs per day, and other practical considerations...

[#8]

Thank you for your answer. I still have some question about the cell batch. Virtually, we could have a endless  batches of cells,  how could we make sure that the production process still work ? 

[#9]

I'm confused by the following:

"but I don't think we could adjust the process since it is unknown what make some stem cells have a better yield than others" And related:

"how could we make sure that the production process still work"

Why don't you try to understand why there is batch to batch variation of cells?

It seems to me that you could/should be trying to figure out what components of variation have the greatest leverage BEFORE starting your experimentation.  Which sources of variation are greatest?  Within or between batch of cells?  You could do sampling (components of variation studies, hierarchical or systematic) to determine the consistency and size of the multiple components (and evaluate your measurement systems against these components) which would be extremely helpful in determining how to handle these sources in an experiment.

[#10]

I'm going to reply in a different way compared to @Phil_Kay and @statman whose technical questions, answers and advice are spot on. If you have the wherewithal, primarily since you are a self professed 'beginner' at DOE, I'd hire a consultant to guide you through this process. There are 1001 issues that have and will rise up and Phil and Statman have just scratched the surface in compact Forum such as this. For example, what if something goes 'wrong' wrt to the 'plan' during experimental conduct? Or measurement of responses? And once you get there, you'll have to start a new thread for analysis. This is not the place to get a speedy answer in real time. A competent consultant with experience in bioassay work can explore lots of issues with you and provide guidance to your questions as well.