Regression analysis of factorially designed trials--a logical approach to protein crystallization

The specific composition of the mother liquor that induces the formation of protein crystal is usually quoted as the 'best condition' for crystallization. However, very little is described about how these conditions were determined. These missing and non-reported details would certainly lead to a better understanding of why a specific purified protein does not crystallize, and why crystallization seems to be a totally unpredictable enterprise. In this paper, it is shown that a simple regression model applied to a set of factorially designed experiments (comprising successful and failed experiments) is a useful tool for the analysis of crystallization outcomes. The applications of this approach to the proteins lysozyme and calmodulin yielded results that did not substantially differ from published accounts. When this procedure is applied to myelin basic protein (MBP) and to a MBP and calmodulin complex no crystals formed, which would suggest that neither is easy to crystallize. In summary, factorial design and regression analysis can be an important methodological approach for crystallizing proteins.