|
Abstract:
|
Gene expression data frequently contain missing values, however, most downstream analyses for microarray experiments require complete data. Many methods have been proposed to estimate missing values, but the specific conditions for which each method is preferred remains largely unclear. In this report we describe an extensive evaluation of current imputation methods on multiple types of microarray experiments. We found that the success of each method depends on the underlying "complexity" of the expression data, i.e. the difficultly in mapping the gene expression matrix to a lower-dimensional subspace. We develop an entropy measure to quantify this complexity, and introduce two complementary selection schemes, entropy-based selection (EBS) and self-training selection (STS), for determining the most appropriate imputation method for any given dataset.
|
