Abstract
This study presents an effective method of blindly classifying large amounts of gene expression data into biologically meaningful groups using a combination of independent component analysis (ICA) and clustering techniques. Specifically, we show that the genes can be classified blindly into several groups based solely on their expression profiles. These groups have a very close correspondence with benchmarks obtained by studies using domain knowledge. These results suggest that ICA can be a very useful pre-processing tool in blind gene classification, rather than using the resulting sources as the final model profiles. © 2005 IEEE.