DOE for serum cell assay optimization

[#1]

Hello I am trying to optimise a cell based assay (process to culture T-cells) to work out combinations of factors (continuous i.e seeding density or process durarion) and levels optimal to lead to highest impact on my list of effects between my two categorical factors ( Serum A and Serum B)

Which DOE and analysis would you recommend?

I can opt for having as little as 2-3 factors and as high as 8-9 factors depending on how much parameters I want to test.

My idea was to run a DOE and then look for those parameters with significant interaction to my Categorical Factor (serum) to work out which conditions are most likely to impact my readouts based on the Serum you use

Thanks in advance!

[#2]

Hi @DRC92,

Welcome in the Community !

The choice and analysis of DoE really depend on your prior knowledge about the system, and the number of factors, constraints, randomization restrictions, noise/signal ratio, ...

You mentioned that you may have as little as 2-3 factors and as high as 8-9 factors. Do you already know the influences and importances of all the possible factors in your system ?

If not, I would recommend starting with a Screening design, to detect important effects and factors in your system.

Once these important effects identified, you can then augment your design to add some points to refine your model (and increase predictivity) and optimize your response(s).

Your initial screening design should include as much factors as possible with large factors range, to identify which factor(s) are important and active in your system.

• Definitive Screening Design may be an interesting choice in the absence of factors constraints (and for 5+ continuous factors, and limited number of 2-levels categorical factors), as it is a powerful 3-levels screening design which can detect main effects, interactions and quadratic effects, for a very limited experimental budget. Having 3-levels for continuous factors help avoid "binary" responses with very high or very low values, or using only "absence/presence" settings for the high and low levels of the factors.
• Custom design (D/A-Optimal screening designs) may also be an interesting alternative, enabling to include various factors types, constraints, and randomization restrictions (for example if you need to setup your plate experiments with a defined number of experiments by row and columns), with full flexibility on the assumed model and what you want to detect. 

I hope this first answer will help you,

[#3]

There is no way to know the best á priori.  My suggestion is to create multiple DOE options (design multiple experiments).  For each experiment, list what potential knowledge you will gain (e.g., what effects can be estimated, what order of model effects can be estimated, what is confounded and what is restricted) and weigh this against the resources you have available.  Predict ALL possible outcomes of each experiment, then pick one and run it knowing this may be the first in a series of experiments.