|
Abstract:
|
High dimensionality and nonlinearity are two important features in large scale, complex genomic data. The first feature has drawn a lot more attention from many researchers. In microarray gene expression studies, various clustering, dimension reduction, and variable selection procedures have been successfully applied. Gene clustering is often accomplished by measuring the similarity between gene expression profiles. The commonly used similarity measure, correlation coefficient, is inadequate for discovering nonlinearity features. Recently, we developed an online bio-computing system based on the new statistical concept of liquid association (LA). LA measures the change in correlation between a pair of variables as mediated by a third variable. The statistical properties of LA and its ability to exploit both the dimensionality and nonlinearity in complex genomic data will be discussed.
|
