Analysis of imputation algorithms for microarray gene expression data

Microarray technology makes it possible to measure expression level of thousands of genes simultaneously in an efficient and inexpensive manner. However, due to various complexities in processing microarrays, expression information of various genes may be missing due to unreliable measurements. The occurrence of missing values in gene expression data can adversely affect downstream analyses such as clustering, dimensionality reduction etc. Different algorithms have been developed to estimate the missing values in different data sets and none of these algorithm works well with all the data sets. In this work, we explore the possible application of Mutual Nearest Neighbor (MNN) algorithm to impute the missing values, which shows comparable results with other well know imputation algorithms. We also have explored five different methods for missing value imputation namely Row Average Imputation, Mean Imputation, Median Imputation, k-Nearest Neighbor Imputation and combination of kNN based feature selection (kNNFS) and kNN-based imputation. The experiments are carried out on very high dimensional gene expression data such as Notterman Carcinoma and Notterman Adenocarcinoma data and the results are illustrated.