• First, it increases our ability to discriminate between different tumors in ways that complement the standard criteria applied by pathologists to characterize tumors, such as histological appearance, cytogenetic markers, and expression of specific marker proteins. Once distinguishing signatures for different tumor types (e.g., tumor A versus tumor B) are defined using known samples, the expression pattern of unknown tumor samples can then be compared with the expression signatures for tumor A and tumor B and classified as A-like, B-like, or neither, depending on how well their expression profiles match the signatures of A and B. Pathologists have used expression profiling to make difficult distinctions between tumors that require very different management approaches. These include distinguishing large B-cell lymphoma from Burkitt lymphoma, differentiating primary lung cancers from squamous cell carcinomas of the head and neck metastatic to lung, and identifying the tissue of origin of a cryptic primary tumor whose metastasis gives too little information to allow its classification.

• Second, different signatures may be found to correlate with known clinical outcomes, such as prognosis, response to therapy, or any other outcome of interest. If validated, such signatures can be applied prospectively to help guide therapy in newly diagnosed patients.

• Finally, for basic research, clustering may reveal previously unsuspected connections of functional importance among genes involved in a disease process.