Gene expression data matrices often contain missing expression values. Inthis paper, we describe a new algorithm, named improved fixed rankapproximation algorithm (IFRAA), for missing values estimations of the largegene expression data matrices. We compare the present algorithm with the twoexisting and widely used methods for reconstructing missing entries for DNAmicroarray gene expression data: the Bayesian principal component analysis(BPCA) and the local least squares imputation method (LLS). The threealgorithms were applied to four microarray data sets and two synthetic low-rankdata matrices. Certain percentages of the elements of these data sets wererandomly deleted, and the three algorithms were used to recover them. Inconclusion IFRAA appears to be the most reliable and accurate approach forrecovering missing DNA microarray gene expression data, or any other noisy datamatrices that are effectively low rank.
展开▼
