We wish to consider the joint effects of genes in determining classification rules for discriminating tumors. There are two assumptions that drive our proposed methodology. First, we assume that the joint effects of multiple genes must be considered in discriminating classes of disease. Recently, much attention has been given to the finding that a 70-gene signature can predict breast cancer survival (van't Veer et al [8]; van de Vijver et al [9]). However, most such gene signatures have been constructed using univariate methods. It seems reasonable to consider joint models, as genes are correlated because of their mutual involvement in disease pathways.
