Experimental Strategy for High Throughput Materials Development
- PDF / 5,785,995 Bytes
- 5 Pages / 612 x 792 pts (letter) Page_size
- 13 Downloads / 218 Views
Experimental Strategy for High Throughput Materials Development James N. Cawse Cawse and Effect LLC 132 Kittredge Rd Pittsfield, MA 01201 USA ABSTRACT Successful experimental strategy in the high throughput world requires coordination of the statistical planning of the experimental program with the optimization of the workflow for efficient use of robotic workstation resources. INTRODUCTION Development of an experimental strategy for a high throughput experimental program requires a different level of planning than a conventional Design of Experiments (DOE) experiment. Typically, the numbers of factors, levels, possible interactions, and constraints are much greater than in the experiments discussed in standard statistics textbooks such as Montgomery [1] or Box, Hunter and Hunter [2]. Systematic multistage experimentation is required to properly investigate these vast experimental spaces [3]. One strategic element that does not arise at all in the statistical planning is optimizing of the robotic apparatus that actually performs the experiments. THEORY Almost all high-throughput experimentation uses one or another type of automated workstation (Figures 1). These typically have a fixed format of sample positions and one or more robotic tools to add, fill, mix, cap, agitate, and heat a set of reactions. These fixed-format systems give rise to what may be called the “Procrustes Problem” (Figure 2). A statistically optimal set of experiments generally will not fit neatly into the fixed format of the robotic workstation. Careful decisions on the selection or abandonment of statistical principles are required in order to make meaningful set of experiments that will operate well in the robotic workflow.
Figure 1. Automated workstations. Left, www.chemspeed.com © 2011 Chemspeed, used by permission; right, www.freeslate.com ©2011 Freeslate, used by permission.
Figure 2. In Greek mythology Procrustes was a rogue smith and bandit from Attica who physically attacked people by stretching them or cutting off their legs, so as to force them to fit the size of an iron bed. http://en.wikipedia.org/wiki/Procrustes . Image © 1987 Chelsea House Publishers, used by permission.
. DISCUSSION Consider a fairly typical (and actually rather small) search for new catalysts (Table 1). The experiments are to be run in a 48-well workstation, and a budget of 100-200 workstation runs is contemplated. This is still less than 1/3 of the all-combinations scenario. The first critical decision to be made is how to select a subset of the all-combinations scenario that will sample the space adequately.
Table 1. Factors and Levels of catalyst search program. Factors Levels Ligands 240 Metals 15 Precursors 2 Cocatalysts 2 Temperatures 2 All Combinations 28,800 One way is to adapt DOE thinking and consider all the different effects and interactions among these factors (Table 2). These can then be prioritized to produce a working model to be used with a D- or I-optimal design to generate an experiment of appropriate size (Table 3). I-optimal designs ar
Data Loading...
