Convex matrix factorization for gene expression analysis

Abstract

A method is proposed for gene expression analysis relying upon convex matrix factorization (CMF). In CMF, one of the matrix factors has a convexity constraint, that is, each row is nonnegative and sums to one, and hence can be interpreted as a probability distribution. This is motivated biologically by expression data resulting from a mixture of different cell types. This thesis investigates implementing CMF with various constraints applied to the expression matrix, and applies the technique to a problem in analysis of the cell cycle and two problems in cancer classification.