Extending Graybill's Legacy to Microarray Data and Beyond
Russell D. Wolfinger, PhD Director of Genomics SAS Institute Inc.
The excitement around microarray data analysis and the accompanying deluge of new research has been a great boon to the statistical profession and provides demonstration of one of our core strengths: the ability to apply, adapt, and extend classical methods to new problems. More than ever, applied statisticians are in high demand by molecular scientists, but are increasingly confronted with an overwhelming number of methods and software for array data.
Some perspective can be drawn from our wise forebears, among whose number is certainly Franklin A. Graybill. His popular papers, books, and even a simple website at http://www.stat.colostate.edu/Facultystaff/graybill.html describe three areas of interest:
· Design of experiments · Linear models · Variance components
One thesis of this talk is that expertise in these areas, potentially more than with any other classical statistical methods, remains largely untapped in the current practice of microarray data analysis. We'll focus on the designed microarray experiment as a principal vehicle for transcriptomic knowledge discovery, and discuss some principles behind good designs. As always, good designs lead to good analyses and optimal information gain from a perforce limited number of runs. An analysis methodology of choice in this regard is the general mixed linear model, and we'll discuss its suitability for both Affymetrix and two-color arrays, as well as it's accommodation of related data sources, both genomic and phenotypic. Time permitting, we'll describe some more advanced mixed models and ideas for future progress.