Multi-domain gating network for classification of cancer cells using gene expression data

Gene expression data (GED) can be indicative of the status of a cell, i.e., healthy or not, type 1 or type 2 of a disease, etc. However, GED differences between cells may be so subtle that most pattern recognition tools can not accurately discriminate them. Toure and Basu (2001) explored the ability of monolithic neural networks and modular neural networks to classify two type of acute leukemia: acute myeloid leukemia (AML) and acute lymphoblastic leukemia(ALL). In this work, we show that modular neural networks are better suited for GED based classification due the high dimensionality and multistructural properties of the input data. A modular network has the ability to examine the data simultaneously in more than one input space. This approach provides more information to the classifier and overcomes various limitations present in the training data