Use of biclustering for missing value imputation in gene expression data

DNA microarray data always contains missing values. As subsequent analysis such as biclustering can only be applied on complete data, these missing values have to be imputed before any biclusters can be detected. Existing imputation methods exploit coherence among expression values in the microarray data. In view that biclustering attempts to find correlated expression values within the data, we propose to combine the missing value imputation and biclustering into a single framework in which the two processes are performed iteratively. In this way, the missing value imputation can improve bicluster analysis and the coherence in detected biclusters can be exploited for better missing value estimation. Experiments have been conducted on artificial datasets and real datasets to verify the effectiveness of the proposed algorithm in reducing estimation errors of missing values.