One-factor-at-a-time (OFAT) experimentation remains common in the pharmaceutical industry, particularly during early development, despite well‑known efficiency limitations compared with design of experiments (DOE). This poster presents a gene therapy case study illustrating how JMP supported a transition from OFAT to a mixture DOE strategy through visualization‑driven design choices and intuitive exploration of experimental space.

The application involved plasmid components with proportions constrained to sum to unity, making the system inherently mixture‑based. JMP’s ternary plots were used to clearly visualize the feasible formulation space, establishing a shared understanding of the mixture structure and enabling more effective scientific discussion around experimental strategy.

A space filling mixture design was implemented in JMP to explore the formulation region. Unlike traditional optimal designs that often emphasize boundary points, the space filling approach provided even coverage of the design space. This coverage aligned well with experimental intuition, increased confidence in the design, and supported a shift away from OFAT.

This case study demonstrates how JMP’s visualization tools and space filling mixture designs can support practical adoption of DOE by aligning statistical rigor with how scientists explore and reason about experimental space in gene therapy applications.

 

 

Hello, everyone. I'm Yang, and I'm a statistician with Johnson & Johnson. I will be sharing a story about how we shifted from a long history of using OFAT to applying space-filling design for the mixture of plasmids in a gene therapy application. This is joint work with Lisa and Allison, the scientists on the project. I am excited to share the turning points of that mindset shift with the help of easy-to-use JMP features.

Here is the project background. The scientist team is working on process optimization for a gene therapy product. The goal is to maximize yield and minimize impurities. The product is made from three plasmids, and I will call them the three ingredients of the recipe. The team had already completed 24 runs when they reached out to me. For a long time, they studied the ingredients one at a time in conjunction with scientific knowledge, treating each as an independent input, both in the experiments and in the analysis. The three ingredients must add up to 100%. If you add more of one, you get less of the others. If you include all three ingredients in the model, you get highly inflated uncertainty and unstable, unreliable results, as shown on the right of the slide.

The first step was not to run more experiments. It was to look at the data they already had in the right way. First, a note about the ratios. In the scientist data sheet, a recipe like 1:1:1 means the three ingredients are equal. In mixture terms, that same recipe is 33%, 33%, and 33%. We need the percent version in the mixture framework. Additionally, on the left is the old way of visualizing the data: three ingredients viewed separately, without accounting for the correlation between them. It tells us something, but it is hard to see what you have tested and just as hard to see what you have missed. On the right is the same data converted into the mixture framework and shown on a ternary plot in JMP. Each corner means 100% of one ingredient. Every point inside is a mix that adds up to 100%. The triangle shows every recipe you could make.

Now you can see it clearly that the historical runs explored a relatively narrow region with large empty area around it. Same data, a very different picture. After visualizing the gaps based on historical data, we can then define the target region for the next DOE. The blue frame marks what was explored. Within part of it, the earlier results showed a promising trend — promising, but not proven. The purple frame filled in white is the region the scientists wanted to explore next, extending out from that trend. This is the most important step. Before, the team only had a general idea of where to look. Now, they had a clear area on the picture that they could point to and see with their own eyes. Being able to sit with the scientists and explore this plot live in JMP is what made the next discussion easy. Knowing where to look, the next question is where to put our new experiments inside that area.

We compared two DOE plans: optimal mixture versus space-filling mixture design. The optimal design is efficient, but it puts most points near the edges of the area and around the center. That does not give enough assurance to scientists who are used to getting information only from the actual settings they explored. The space-filling design solves exactly that. It spreads the points evenly, so the scientists could see at a glance that the entire region of interest was covered, and that is why we chose it. That assurance is what moved them away from OFAT.

The final design had 23 experimental runs: 18 points from the design, 4 replicates at the corners… Sorry, replicates at the four corners added at the scientists' request, and one standard recipe for comparison. The space-filling design has been executed, and we are now collecting the data and performing the analysis. Here are the impacts from both the technical side and scientific side. On the technical side, we turned the old OFAT data into a solid experiment plan. The ternary plots in JMP showed us what had never been tested and gave us a clear idea of where to go next.

More important is the scientific impact. The team now has a simple way to picture their process, which makes DOE easier to discuss and apply than before. After this project, the scientist team requested DOE training and specifically asked for mixture design. The real result was not a better experiment; it was a better way to think about the problem. The visuals and the design tools in JMP did the hard work of making mixture thinking easy to see and easy to trust. That is what won the scientists over. Thank you all for your attention.

0 Comments
Presented At Discovery Summit 2026

Presenters

Skill level

Intermediate
  • Beginner
  • Intermediate
  • Advanced

Files

Published on ‎07-15-2026 03:39 PM by Community Manager Community Manager | Updated on ‎10-30-2026 03:35 PM

One-factor-at-a-time (OFAT) experimentation remains common in the pharmaceutical industry, particularly during early development, despite well‑known efficiency limitations compared with design of experiments (DOE). This poster presents a gene therapy case study illustrating how JMP supported a transition from OFAT to a mixture DOE strategy through visualization‑driven design choices and intuitive exploration of experimental space.

The application involved plasmid components with proportions constrained to sum to unity, making the system inherently mixture‑based. JMP’s ternary plots were used to clearly visualize the feasible formulation space, establishing a shared understanding of the mixture structure and enabling more effective scientific discussion around experimental strategy.

A space filling mixture design was implemented in JMP to explore the formulation region. Unlike traditional optimal designs that often emphasize boundary points, the space filling approach provided even coverage of the design space. This coverage aligned well with experimental intuition, increased confidence in the design, and supported a shift away from OFAT.

This case study demonstrates how JMP’s visualization tools and space filling mixture designs can support practical adoption of DOE by aligning statistical rigor with how scientists explore and reason about experimental space in gene therapy applications.

 

 

Hello, everyone. I'm Yang, and I'm a statistician with Johnson & Johnson. I will be sharing a story about how we shifted from a long history of using OFAT to applying space-filling design for the mixture of plasmids in a gene therapy application. This is joint work with Lisa and Allison, the scientists on the project. I am excited to share the turning points of that mindset shift with the help of easy-to-use JMP features.

Here is the project background. The scientist team is working on process optimization for a gene therapy product. The goal is to maximize yield and minimize impurities. The product is made from three plasmids, and I will call them the three ingredients of the recipe. The team had already completed 24 runs when they reached out to me. For a long time, they studied the ingredients one at a time in conjunction with scientific knowledge, treating each as an independent input, both in the experiments and in the analysis. The three ingredients must add up to 100%. If you add more of one, you get less of the others. If you include all three ingredients in the model, you get highly inflated uncertainty and unstable, unreliable results, as shown on the right of the slide.

The first step was not to run more experiments. It was to look at the data they already had in the right way. First, a note about the ratios. In the scientist data sheet, a recipe like 1:1:1 means the three ingredients are equal. In mixture terms, that same recipe is 33%, 33%, and 33%. We need the percent version in the mixture framework. Additionally, on the left is the old way of visualizing the data: three ingredients viewed separately, without accounting for the correlation between them. It tells us something, but it is hard to see what you have tested and just as hard to see what you have missed. On the right is the same data converted into the mixture framework and shown on a ternary plot in JMP. Each corner means 100% of one ingredient. Every point inside is a mix that adds up to 100%. The triangle shows every recipe you could make.

Now you can see it clearly that the historical runs explored a relatively narrow region with large empty area around it. Same data, a very different picture. After visualizing the gaps based on historical data, we can then define the target region for the next DOE. The blue frame marks what was explored. Within part of it, the earlier results showed a promising trend — promising, but not proven. The purple frame filled in white is the region the scientists wanted to explore next, extending out from that trend. This is the most important step. Before, the team only had a general idea of where to look. Now, they had a clear area on the picture that they could point to and see with their own eyes. Being able to sit with the scientists and explore this plot live in JMP is what made the next discussion easy. Knowing where to look, the next question is where to put our new experiments inside that area.

We compared two DOE plans: optimal mixture versus space-filling mixture design. The optimal design is efficient, but it puts most points near the edges of the area and around the center. That does not give enough assurance to scientists who are used to getting information only from the actual settings they explored. The space-filling design solves exactly that. It spreads the points evenly, so the scientists could see at a glance that the entire region of interest was covered, and that is why we chose it. That assurance is what moved them away from OFAT.

The final design had 23 experimental runs: 18 points from the design, 4 replicates at the corners… Sorry, replicates at the four corners added at the scientists' request, and one standard recipe for comparison. The space-filling design has been executed, and we are now collecting the data and performing the analysis. Here are the impacts from both the technical side and scientific side. On the technical side, we turned the old OFAT data into a solid experiment plan. The ternary plots in JMP showed us what had never been tested and gave us a clear idea of where to go next.

More important is the scientific impact. The team now has a simple way to picture their process, which makes DOE easier to discuss and apply than before. After this project, the scientist team requested DOE training and specifically asked for mixture design. The real result was not a better experiment; it was a better way to think about the problem. The visuals and the design tools in JMP did the hard work of making mixture thinking easy to see and easy to trust. That is what won the scientists over. Thank you all for your attention.



Start:
Tue, Oct 20, 2026 04:30 PM EDT
End:
Tue, Oct 20, 2026 05:15 PM EDT
Ped 3
Attachments
0 Kudos