Bayesian Machine-Learning Methods for Tumor Classification Using Gene Expression Data
Precise classification of tumors is crucial for cancer diagnosis and treatment. The ability to target specific therapies to pathogenetically distinct tumor types is very important for cancer treatment because it maximizes efficacy and minimizes toxicity . Diagnostic pathology has traditionally relied on macroand microscopic histology and tumor morphology as the basis for tumor classification. The downside of it, however, is the inability to discriminate among tumors with similar histopathologic features, as these features vary in clinical course and in response to treatment. This justifies the growing interest in changing the basis of tumor classification from morphologic to molecular, using microarrays which provide expression measurements for thousands of genes simultaneously [35, 10]. The idea is to carry out classification on the basis of different expression patterns. Several studies using microarrays to profile colon, breast and other tumors have demonstrated the potential power of this idea [2, 21]. Gene expression profiles may offer more information than, and provide an alternative to, morphology-based tumor classification systems. The introduction of microarray technology has raised a variety of statistical issues, ranging from low-level issues such as preprocessing and normalization of the data [44, 53] to differential expression of genes in experimental conditions
[12, 24]. In this chapter, we will be focusing on the problem of classification based on microarray data.