Hi all,
This might be really basic, but being relatively new to JMP I have struggled quite a bit with setting up a DOE that will allow me to evaluate a number of factors including the column and row number (and their interaction) in a 24-well plate. My issue was, that when I tried to include columns and rows in the design, some positions might appear twice and some not at all. If I tried to make a design including all other factors and then manually add positions to the resulting table and to the model the VIF's (i.e. co-linearity) were sometimes really high. This might also be an issue for 96-well assays although probably less so because replicates would presumably be spread out more by coincidence.
After struggling with this for a while I have apparently found a solution which is to include columns and rows in the design as covariate factors. Specifically, I made a data table with all row/columns combinations and with this table active selected custom design from the DOE menu. This allowed me to include rows and columns as covariate factors in the actual design and it appears that the randomization now takes the positions into account and minimizes the colinearity in the design table.
Note, that if your design involves more plates than one, you need to copy the 24 different row/column combinations as many times as the number of plates you wish to include. The attached file "24_plate-9_runs" is simply 9x24 positions that I used as input for the DOE found in "24DWP_DOE-example". Here, I included 3 plates x 3 days as "Hard" and "Very Hard" to change factors, respectively, resulting in a design that allows me to evaluate day-to-day, plate-to-plate and well-to-well variation as well as any systematic effects of column or row number. Note that in the attached exampe I have manually sorted the well positions to allow me to dose enzyme using a liquid handler.
Please let me know if there is an easier way to achieve this or if for some reason my approach is flawed in terms of the quality of the resulting design.
Great, it looks like it works! Thanks for the input:-)
Johan
Hi
I am also fairly new to DoE and had a similar issue. I found that using the D-optimal Design as the optimality criterion (Custom design (red triangle) --> Optimality criterion) gave me the accurate distribution when I added the rows and columns as categorical factors. I sometimes had to increase the calculation time to ensure the correct distribution (Custom design (red triangle) --> Design search time). 60 s was enough in my case.
I hope this helps,
Katrine
