The estimation procedure described in this paper allows the joint estimation of multivariate gene effects on the response (class label). The approach described here could be generalized by fitting more nonlinear gene effects in the estimation algorithm or by including higher-order interactions between genes. Another generalization is to perform a clustering of the genes and to enter the cluster averages as covariates in the model. Such an approach was taken by Hastie et al [20] and Tibshirani et al [21].
