Patient response datasets: Challenges and opportunities

As the field of bioinformatics has grown in importance, more and more studies have investigated the use of gene microarray datasets to understand cancer. Although much of this research has focused on which genes are differently-expressed between cancerous and non-cancerous tissues, an equally important question is which genes are most useful for predicting the success of cancer treatment. How well a patient will respond to a given treatment depends on the specifics of their cancer, and biopsies alone cannot detect genetic markers; thus, gene chips are an increasingly valuable research tool in this field. The problem of identifying which gene markers are predictive of successful response to treatment differs from more general cancer-identification and cancer-classification problems due to the expected similarities among all patients (since they share a cancer type), and thus it is important to understand this collection of datasets as a group separate from other cancer-related microarray datasets. The present work surveys research using gene microarray datasets for the task of patient response prediction, in particular research which uses this data to discover the best genes for predicting patient response. We discuss the methods and procedures of the surveyed papers and which approaches were used for gene selection, and present ideas and strategies for future work which further explores how to best identify gene signatures which can be used clinically to help select the best treatment for a given cancer.