Gene expression patterns and cancer classification: a self-adaptive and incremental neural approach

The automated interpretation of data originating from the human genome may play a crucial role in cancer treatment. In this paper, a new computational approach to the discovery and analysis of gene expression patterns is presented and applied to the recognition of B-cell malignancies as a test set. Using cDNA microarray data, an unsupervised and self-adaptive neural network model known as "growing cell structures" is able to identify normal and diffuse large B-cell lymphoma (DLBCL) patients. Furthermore, it discovers the distinction between patients with molecularly distinct types of DLBCL without previous knowledge of those subclasses.