EMMA: An EM-based Imputation Technique for Handling Missing Sample-Values in Microarray Expression Profiles

Corresponding author Abstract Data with missing sample-values are quite common in many microarray expression profiles. The outcome of the analysis of these microarray data mostly depends on the quality of underlying data. In fact, without complete data, most computational approaches fail to deliver the expected performance. So, filling out missing values in the microarray, if any, is a prerequisite for successful data analysis. In this paper, we propose an ExpectationMaximization (EM) inspired approach that handles a substantial amount of missing values with the objective of improving imputation accuracy. Here, each missing samplevalue is iteratively filled out using an updater (predictor) constructed from the known values and predicted values from the previous iteration. We demonstrate that our approach significantly outperforms some standard methods in terms of treating missing values, and shows robustness in increasing levels of missing rates.