Sparse regularized discriminant analysis with application to microarrays

For cancer prediction using large-scale gene expression data, it often helps to incorporate gene interactions in the model. However it is not straightforward to simultaneously select important genes while modeling gene interactions. Some heuristic approaches have been proposed in the literature. In this paper, we study a unified modeling approach based on the ℓ1 penalized likelihood estimation that can simultaneously select important genes and model gene interactions. We will illustrate its competitive performance through simulation studies and applications to public microarray data.

Figure optionsDownload as PowerPoint slideHighlights
► A unified modeling approach that simultaneously analyzes gene interactions and selects important genes for improved prediction of microarrays.
► Very efficient computational algorithms developed for model estimation and selection.
► Demonstrate the very competitive performance of the proposed method.