Design of Experiment (DoE)
One of the most valuable tools I've acquired in my professional career is the Design and Analysis of Experiments (DoE). DoE is the science of efficient research, enabling researchers to use statistical methods to optimize experimental design and maximize insights with minimal sampling.
My presentation at the 2023 Society of In Vitro Biology Design of Experiment Workshop
Is expensive commercial Design of Experiment (DoE) software necessary for efficient research in multidimensional spaces? At the 2023 SIVB DoE Workshop, I argued that the free statistical software R is an excellent option for practicing DoE. While the initial learning curve for R may be steeper compared to commercial DoE software, mastering R can benefit professionals in multiple aspects of their careers beyond research.
Free DoE: A Free Design of Experiment App Built with Shiny
Interested in learning the science of experimental design without investing in costly commercial DoE software or learning a programming language? This free DoE app is for you. Leveraging Shiny to make R’s experimental design packages accessible to non-programmers, this app introduces practitioners to DoE. While it may not have all the features of commercial software, it allows users to design and analyze important optimization and screening experiments at no cost.
Fractional Factorial Designs
Often, the most challenging aspect of process improvement is determining where to focus improvement efforts. Optimization designs typically require numerous iterations, making it essential to identify the key factors and interactions that significantly impact the system's response prior to use. This is where screening designs come into play. The Fractional Factorial Design enables researchers to explore an extensive design space with minimal runs while also detecting interactions.
Response Surface Methodology
Once you have identified the critical factors that control your process, it is time to begin optimization. Response Surface Methodology (RSM) allows researchers to build multiple linear regression models, which can predict responses based on specific input values. If the design space can be adequately modeled using regression, the predictive model can be used to precisely control the expected output. For processes that influence product cost, Response Surface Methodology can be a powerful tool for improving profitability.
Mixture Designs
Mixture models are an incredibly useful class of designs, particularly when there is an imposed limitation in the system. Given this limitation, an experimenter can evaluate the system as if one less variable is involved. Knowing the level of one variable and the physical limit allows us to deduce the level of the second variable. In this blog post, I will demonstrate how I have used mixture designs to optimize the mineral nutrition for plant tissue culture media.