Computational Approaches to Cancer Classification

Abstract

Although cancer classification and treatment have improved greatly over the past 30 years, there was until recently no general molecular strategy for identifying new cancer classes or for assigning tumors to known classes. However, in the mid-1990s, the advent of microarray technologies that measure the activity of thousands of genes in a cell population raised the possibility of studying cancer at the molecular level to find new clues to patient prognosis and treatment.

In this talk, we will introduce the technology that has enabled this revolution in cancer research and will discuss computational approaches to the molecular classification of cancer. We will begin by describing our early work classifying different types of human leukemia, a result that proved the feasibility of molecular cancer classification despite the problem's potential computational and biological intractability. We will discuss other computational approaches to this multivariate machine learning problem, and raise the question of whether more sophisticated learning techniques are necessarily better for this particular class of problems. We illustrate the strengths of relatively simple prediction techniques in more recent work, including diagnosing renal cancer from blood samples and predicting patient responses to drug treatment.

Our results demonstrate the feasibility of cancer classification based solely on gene expression monitoring, and suggest a general strategy for discovering and predicting cancer classes for other types of cancer. They also illustrate the practical impact that computational biology can have on modern medicine.