Ontology Assisted Unsupervised Clustering for Interpreting Microarray Gene Expression Profiles

Analysis of microarray gene expression profiles has been a routine step in finding the relationships between gene and associated disease. While it is possible to measure thousands of gene activities simultaneously using microarray technology, finding functionally similar genes or gene patterns under a given set of experimental conditions remains challenging. Although unsupervised clustering has been the basic composition of microarray data analysis, it neither provides the proof of gene-gene relationship nor the best possible grouping by gene function. Therefore, in this work, we have developed a framework that combined gene ontology with unsupervised clustering to infer expression profiles of Saccharomyces cervisiae, measured in cultured yeast, and identified key mitochondrial genes that are likely involved in diauxic-shift. By combining gene ontology with two unsupervised clustering techniques, a set of regulatory genes were identified showing the relationships between mitochondrial pathways of ATP synthesis and glucose fermentation in cultured yeast. It was demonstrated that ontology-based cluster interpretation provides a powerful tool for exploratory analysis of microarray time series data, and improves the cluster interpretation during gene-pathway mapping. The proposed method can be applied to study genome-scale metabolic regulation in other cell types.