Investigation of the effects of imputation methods for gene regulatory networks modelling using dynamic bayesian networks

Abstract

DNA microarray technology plays an important role in advancing the analysis of gene expression and gene functions. However, gene expression data often contain missing values, which cause problems as most of the analysis methods of gene expression data require a complete matrix. Several missing value imputation methods have been developed to overcome the problems. In this paper, effects of the missing value imputation methods in modeling of gene regulatory network are investigated. Three missing value imputation methods are used, which are k-Nearest Neighbor (kNN), Iterated Local Least Squares (ILLsimpute), and Fixed Rank Approximation Algorithm (FRAA). Dataset used in this paper is E. coli. The results suggest that the performance of each missing value imputation method is influenced by the percentage and distribution of the missing values in the dataset, which subsequently affect the modeling of gene regulatory network using Dynamic Bayesian network.