I recently spent two valuable days at the Freeslate Forum in San Francisco. Two of the case studies were on the use of high throughput tools in more academic settings, by Emory Chan of The Molecular Foundry at Lawrence Berkeley National Laboratory and by Scott Virgil at the Center for Catalysis and Chemical Synthesis (3CS) at Cal Tech. Both talks emphasized the use of Freeslate equipment to allow professors, students and visiting projects to rapidly and thoroughly get results on their chemistry. Projects varied from nanoparticle synthesis to organic synthesis, catalysis, kinetics, and crystallization.
In my talk, a case study on Dow Corning’s development of silicone resin LED phosphors , I emphasized the use of experimental design for high throughput experiments (DOE) to produce the maximum amount of information by selection of a minimal subset of the millions of runs possible in a high throughput project. In the study, I gradually parsed the system from 316,800 runs to 40,500 to a doable 1,512 runs. This required deconstructing a large design into its components and using DOE techniques to select a sparse but useful subset from each component.
I contrasted this with a metathesis catalyst screen done at 3CS which performed all 576 combinations of 12 catalysts, 4 solvents, 3 concentrations, 2 catalyst mol% and 2 temperatures. Although this was quite doable with the Freeslate equipment, I remarked that it was still an overuse of resources. The experiment could easily have been broken down into two smaller components:
- A 48-4un set (qualitative, 12 catalysts x 4 solvents)
- A 5-run set (quantitative fractional factorial in concentration x mol% x temperatures, which would actually give more useful data)
These two sets could be combined to in a 48 x 5 = 240 run design which would give as much or more useful information, using 40% of the available resources!
These remarks led to a lively discussion. One of the key points made by Scott and Emory was that their students and clients just wanted results from their chemistry. If an all-combinations design could be done by the robot without requiring more thinking, it was just fine!
I responded that they were missing a crucial educational opportunity! The level of understanding of DOE in contemporary science and engineering graduates is little better than it was when I got my Ph.D. 40 years ago – abysmal! Given the renewed emphasis on use of experimental design techniques in industry (Lean Six Sigma) and in biotech and pharma (Quality by Design), new scientists and engineers should be robustly encouraged to learn about and use these methods. This applies to high throughput methods as well as the classical techniques.
I would welcome a continued discussion on the appropriate introduction of DOE in the academic high throughput world!