Microarray Design Using the Hilbert-Schmidt Independence Criterion

Justin Bedo

Abstract: This paper explores the design problem of selecting a small subset of clones from a large pool for creation of a microarray plate. A new kernel based unsupervised feature selection method using the Hilbert–Schmidt independence criterion (hsic) is presented and evaluated on three microarray datasets: the Alon colon cancer dataset, the van ’t Veer breast cancer dataset, and a multiclass cancer of unknown primary dataset. The experiments show that subsets selected by the hsic resulted in equivalent or better performance than supervised feature selection, with the added benefit that the subsets are not target specific.