Wednesday, September 26, 2012

Design of Experiments (DoE)

Well, there is a lot of discussion whether to use DoE on method development when working in organic chemistry or not. According to me, you should use it as fast as you can. I mean, the industry has adapted to this quite well and it is now a standard procedure so I do not understand the reluctance from academics to adapt to this powerful tool. There is an excellent article by Torbjörn Lundstedt et al. about DoE (here).

One major reason I like DoE is, I guess, doe to my engineering background. As an engineer, you like graphs, you like correlations and you like computers. DoE offers all of this and assist you in your work. In the paper to which I linked (see above), they address some of the common objections from people reluctant to use DoE. Here is a summary of the normal arguments people use to avoid using DoE (that I have heard my self, there are many more for sure)

“Why Bother When the Results Are Blindingly Obvious?”

Personally I think there is a lot of ego in this kind of statements. Of course you have to have an idea of what will work and what wont work, but given the fact that chemistry is affected by so many parameters (temperature, time, pressure, reagents, rate of addition, catalyst, solvent, concentration, pH, etc), it is worthwhile to try a small screening design at the beginning of any project. At the end, all information might be used and the effect of a given variable can be calculated .

Many people expect DoE to provide miracle results. But, visualize that one-factor-at the time is the counterpart of running with boots, DoE is the counterpart of running with proper running shoes. Having to run is a pain in the *ss no matter the shoes you choose, but you cannot deny that is h*ll more easy to run with Asics Gel-Scout than using a pair of gothic boots.

What If We Miss a Factor? Will My DOE Work Not Be Wasted?

If you miss a factor using DoE, you will probably miss it using the usual approach (one factor at a time). With DoE you can actually use Lack of Fit tests and residual analysis  which may help you find that missing variable (with a little bit of thinking and analysis).

Other stuff you may hear are that the resources are limited. But, using this argument just proves that you have missed the whole point of DoE, it is designed to minimize costs and the amount of experiment that have to be performed in order to find optimal responses. The thing is that you see in beforehand the amount of material that will be "wasted". Changing one-factor-at-the time, you waste while you progress and at the end, you do not realize that the amount of material used just surpassed the amount of material that would be needed to make a screening design and an optimization.

So, I do not have so much more time to use here. I may be wrong in some points, but my advice for them who are new at DoE in method development are (besides reading the papers above)

  • Search the literature. Remember, four weeks in the lab will save you four hours in front of the computer (or library).
  • Start with a screening design. What parameters are important? (usually, temperature and solvent have to be included...)
  • Do not forget to try to minimize the number of variables used. Sometimes using A, B and C (three variables) is just as meaningful as using ratios of these factors (i.e. A/B and C/B, which give you two variables instead of three)
  • When you are done with the screening design and have found the most important variables, is time to optimize. Response surface design is usually the way to go.

Well, the reason I posted this is because of a little bit of frustration I feel due to the almost non-existent use of DoE at academic organic chemistry labs, while this is a method that is a must out in industry. I have been striving to work with it as much as I possibly can, and one day, I hope it will be a tool as frequently used as an analytical HPLC. 

No comments:

Post a Comment