Discovering Gene-Drug Relationships for the Pharmacology of Cancer

he combined analysis of tissue microarray and drug response datasets has the potential of revealing valuable knowledge about the relationships between gene expression and drug activity of tumor cells. However, the amount and the complexity of biological data needs appropriate data mining and machine learning algorithms to uncover possible interesting patterns. In order to identify a suitable profile of cancer patients for revealing the link between gene expression profiles, drug activity responses and type of cancer, a learning framework based on three building blocks is proposed: p-Median based clustering, information gain feature selection and Bayesian Network prediction. The experimental investigation highlights three main findings: (1) the relational clustering approach is able to create groups of cell lines that are highly correlated both in terms of gene expression and drug response; (2) from a biological point of view, the gene selection performed on these clusters allows for the identification of a subset of genes that are strongly involved into several cancer processes; (3) the final prediction of drug responses, by using the patient profile obtained through clustering and gene selection, represents an initial step for predicting potential useful drugs.