Analysis of Gene Expression Profiles and Drug Activity Patterns by Clustering and Bayesian Network Learning

High-throughput genomic analysis provides insight into a complicated biological phenomena. However, the vast amount of data produced from upto-date biological experimental processes needs appropriate data mining techniques to extract useful information. In this paper, we propose a method based on cluster analysis and Bayesian network learning for the molecular pharmacology of cancer. Specifically, the NCI60 dataset is analysed by soft topographic vector quantization (STVQ) for cluster analysis and by Bayesian network learning for dependency analysis. Our results of the cluster analysis show that gene expression profiles are more related to the kind of cancer than to drug activity patterns. Dependency analysis using Bayesian networks reveals some biologically meaningful relationships among gene expression levels, drug activities, and cancer types, suggesting the usefulness of Bayesian network learning as a method for exploratory analysis of high-throughput genomic data.