Discovery and visualization of co-regulated genes relevant to target diseases

In this thesis, we propose to discover co-regulated genes using microarray expression data, as well as providing visualization functionalities for domain experts to study relationships among discovered co-regulated genes. To discover co-regulated genes, we first use existing gene selection methods to select a small portion of genes which are relevant to the target diseases, on which we build an ordered similarity matrix by using nearest neighbor based similarity assessment criteria. We then apply a threshold based clustering algorithm named Spectral Clustering to the matrix to obtain a number of clusters. The genes which are clustered together in one cluster represent a group of co-regulated genes and to visualize them, we use Java Swings as the tool and develop a visualization platform which provides functionalities for domain experts to study relationships between different groups of co-regulated genes; study internal structures within each group of genes, and investigate details of each individual gene and of course for gene function prediction. Results are analyzed based on microarray expression datasets collected from brain tumor, lung cancers and leukemia samples.