Today we're discussing Design of Experiments, or DOE, and the combination of that technique with laboratory automation. The LabMan reached out to a long-time expert in the field, Paul Taylor, Principal Scientist at Boehringer Ingleheim, to discuss the topic. Paul indicates that DOE as an experimental approach was developed by Sir Ronald Fisher beginning in 1919 during his work at the Rothamsted Experimental Station located at Harpenden, Hertfordshire, England. In his role there as a statistician he developed the concept of ANOVA, or Analysis of Variants, which is still used today to determine which factors in an experiment produce significant effects and whether the response is linear or non-linear. In 1925 he published Statistical Methods for Research Workers, followed in 1935 by The Design of Experiments.

DOE is an approach to optimizing a given experimental operation in an iterative, efficient and statistically-driven way, by systematically evaluating the impact of individual experimental variables on the final, measured outcome of the experiment. The approach can be relatively simple for experiments involving just a few variables, or quite complex for multi-variant experiments. In all cases, the idea is to rigorously explore the impact of variables in a well-organized, highly systematic way that minimizes the number of tests to be run and maximizes the statistical relevance of the results.

The rub has always been that it can be a lot of work to do DOE on even a moderately complex problem. It can take significant time to design the array, or table of experiments to sufficiently evaluate all the variables inherent to the problem. This requires knowledge of both the science involved and experimental statistics - usually not resident in the same person. It can then be very tedious and time consuming to run all these experiments. In some cases, there may not be enough supply of experimental components (such as reagents) to run all the desired experiments. In other cases, experimental components may not be sufficiently stable across the span of time necessary for a human to complete the entire planned table of experiments. Many, even most scientists will take intuitive shortcuts rather than go through the DOE tedium. Sometimes this works and other times experimental blind alleys end up taking as much or more time than a carefully planned DOE approach would have.

So, we have a powerful experimental approach that has the disadvantage of being tedious, computationally intensive, involving laborious setup, and consuming both time and supplies. It sounds like the ideal situation for the application of computerized design with automated execution! Specifically, for chemical/biochemical experiments, there are now many choices of automated liquid handling systems that can assume the tedium and complexity of executing a DOE array of experiments, and can do so in a fast and miniaturized way. Computerized power for designing a DOE array of experiments and analyzing the results certainly exists. As Paul points out, the key is then in linking the two into a package that intuitively appeals to the scientist.

Several technology providers have taken a stab at this. Paul has spent a good deal of time working with and helping to develop the Automated Assay Optimization (AAO) package sold by Beckman Coulter for their Biomek FX. Symyx has developed Lab Execution software for chemical process optimization using their Benchtop System. Others have developed "home-grown" versions of similar approaches. Paul feels that for such systems to be successful, they must be capable of translating experimental inquiry in the way a scientist would think to the physical and practical reality of automated system execution. This would basically involve an automated system capable of reading a scientist-generated table of experimental parameters together with a range for each parameter, and translating that into a series of automated methods for exploring the response range for each parameter.

If you want a deeper introduction to DOE, sign up for the ALA short course on the topic at the next LabAutomation conference!

Until next time,

Domo Arigato, Mr. Roboto