Mutual Information Based Extrinsic Similarity for Microarray Analysis

Genes responding similarly to changing conditions are believed to be functionally related. Identification of such functional relations is crucial for annotation of unknown genes as well as the exploration of the underlying regulatory program. Gene expression profiling experiments provide noisy datasets about how cells respond to different experimental conditions. One way of analyzing these datasets is the identification of gene groups with similar expression patterns. A prevailing technique to find gene pairs with correlated expression profiles is to use linear measures like Pearson's correlation coefficient or Euclidean distance. Similar genes are later compiled into a co-expression network to explore the system-level functionality of genes. However, the noise inherent in microarray datasets reduces the sensitivity of these measures and produces many spurious pairs with no real biological relevance. In this paper, we explore an extrinsic way of calculating similarity of two genes based on their relations with other genes. We show that `similar' pairs identified by extrinsic measures overlap better with known biological annotations available in the Gene Ontology database. Our results also indicate that extrinsic measures are useful in enhancing the quality of co-expression networks and their functional subnetworks.