Discussion papers / Sonderforschungsbereich 386 der Ludwig-Maximilians-Universität München 313
Various supervised learning and gene selection methods have been used for cancer diagnosis. Most of these methods do not consider interactions between genes, although this might be interesting biologically and improve classification accuracy. Here we introduce a new CART-based method to discover emerging patterns. Emerging patterns are structures of the form (X1>a1)AND(X2<a2) that have differing frequencies in the considered classes. Interaction structures of this kind are of great interest in cancer research. Moreover, they can be used to define new variables for classification. Using simulated data sets, we show that our method allows the identification of emerging patterns with high efficiency. We also perform classification using two publicly available data sets (leukemia and colon cancer). For each data set, the method allows efficient classification as well as the identification of interesting patterns.
Emerging patterns interactions between variables classification trees supervised learning