The pharma industry is required to develop, launch and deliver safe and efficacious products with minimal cost, time and resources, in a highly regulated environment. Much real-world experience shows that ‘Design of Experiments’ (DOE) is the best way to arrive at the process understanding required to meet these multiple challenges. Consequently, ‘pharma and biotech companies should strive to leverage this paradigm to the full.
DOE is a widely used and practical method for investigating multi-factor opportunity spaces, and JMP provides world-class design and analysis tools via an intuitive user interface. A structured approach to experimentation leads to the efficient and effective collection of data and has a huge range of application. DOE is used uncover or model relationships between inputs, or factors, and an output, or response, intentionally altering the former and observing whether the latter also changes. Actively altering factors in accordance with a pre-defined plan or design is the most effective method for learning, and results in new understanding that one can rely on to drive actions and interventions.
In most real-world scenarios, there are multiple factors, and a design that changes only one at a time is not just slow, but necessarily risks failing to uncover the joint effect of two or more factors. Such interactions are the norm rather than the exception, and finding and exploiting them is often necessary to correctly optimise responses.
Along with a comprehensive library of time-tested traditional DOE designs, JMP alsofeatures innovative custom designs that allow you to tailor your DOE efforts to address specific technological challenges without wasting valuable resources. Once the data is collected, JMP automates the analysis and statistical model building processes, allowing you to quickly visualise the pattern of response, identify active factors, optimise responses, and provide robust solutions.
JMP® DOE options
Sir Ronald A Fisher established the factorial principle, randomisation, replication and blocking as the four cardinal principles of DOE. However, until recently, constructing and analysing a design that followed these principles was essentially a matter of laborious computation. Despite this burden, practitioners have developed a variety of widely used design families that work in specific situations. JMP provides all these classical design types (including full and partial factorials, screening, response surface, mixture and Taguchi). After specifying factors and responses, JMP enables one to select a suitable design from the available options and includes design evaluation tools, such as prediction variance profiles and FDS plots, to help validate selection prior to allocating any resources. After the experimental runs in the design are complete, analysis is simplified by JMP scripts automatically placed into the data table used to collect the results.
Split plot designs – Experiments with difficult-to-change factors
In many practical situations, factors that are hard to change can throw the analysis off. As mentioned above, randomisation is a key principle of DOE. However, for reasons of cost or convenience, hard-to-change factors cannot be fully randomised. A split plot design is the correct approach in such situations, and JMP can generate split-plot, split-split, and strip-strip designs that are optimal, also allowing the inclusion of covariates. JMP automatically creates a design that respects the restricted randomisation and includes a script for the relevant Random-Effect restricted Maximum Likelihood (REML) model in the data table detailing the experimental runs.
Space-filling,choice and life-testing designs
Even when there is no intrinsic variability in the response, DOE is useful for efficiently investigating many factors. JMP provides space-filling designs that are often analysed using a Gaussian Process smoother to generate a surrogate model with low prediction bias and variation. JMP can also generate and analyse choice designs in which consumers or users are asked to rank options, with or without regard to cost. JMP also provides designs for Accelerated Life Testing and non-linear models.
Experience the power of customised experimentation—Custom designs
Traditional designs use a design library, so there is a risk that no pre-existing design will fit technological problem well. Given a specification of one’s factors, their levels, and one’s prior expectation of the form of the response, JMP allows one to overcome this pitfall by using a Custom Design. A Custom Design is tailored to answer the specific questions of interest, rather than forcing one’s problem to conform to a textbook design.
Custom Designs always make the best use of one’s experimental budget, and can include continuous, multi-level categorical and mixture factors. One can also specify if some factors are hard or very hard to change so that the required split-plot, split-split and strip-strip designs are automatically generated.
As one develops his/her design, he/she can include factor constraints, specify the desired model effects and interactions, and include centre points and or replicate runs as needed. By doing sample size and power calculations and visualising alias structures within the design process, one can determine if their chosen experiment is likely to be effective in practice.
Scan all possibilities with definitive screening designs
Screening designs aim to find the ‘vital few’ factors amongst the ‘trivial many.’ But, by using a traditional screening design, one may miss the fact that the influence of a factor on the response is substantially curved, and then erroneously remove this factor from further consideration. The definitive screening design is a new design type provided by JMP that aims to reduce this risk, while still keeping one’s experimental budget manageable.
Evaluate, compare and augment designs
JMP provides a complete set of design diagnostics that allows one to assess if the experiment they are planning is likely to succeed. JMP lets one look at the prediction variance over the factor space and precision of one’s estimates, as well as any aliasing or inefficiency in one’s design. Using Compare Design, one can study tradeoffs based on the number of runs used, the statistical power, the fraction of design space covered, and other important metrics. Moreover, if you want to take a sequential approach, Augment Design allows you to quickly build on a completed experiment with new runs without sacrifising any of the data you have previously acquired.
Resources
White Paper – Optimizing Processes with Design of Experiments
https://www.jmp.com/en_us/whitepapers/jmp/optimizing-processes-with-doe.html
Video – Design of Experiments for Pharmaceutical and Biotech Manufacturing
In this on-demand webinar, learn about JMP’s DOE advantage, and how to use it to improve processes and products. A case study shows key concepts, including how to plan a designed experiment in JMP and interpret the results.
Download your free trial of JMP®
Used by hundreds of thousands of data explorers worldwide, JMP data analysis software reveals insights that raw tables of numbers or static graphs tend to hide. Get more out of your data by downloading a free, fully functional 30day trial now
Thanks for sharing such an informative piece of content.
Great article here , Good article.