Biclustering Gene Expression Profiles by Alternately Sorting with Weighted Correlated Coefficient

This paper proposes a framework for biclustering gene expression profiles. The framework applies dominant set approach to create sets of sorting vectors. With these sorting vectors, we iteratively sort and transpose the gene expression data. Weighted correlation coefficient is used to measure the similarity in the gene level and the condition level. The weights are assigned according to the similarity measures in the previous level. We refine and update the weights of our similarity measurement in each iteration. This enables us to concentrate on measuring the similarity of relevant features during the biclustering process. In this way, a highly correlated bicluster could be easily located. We have applied this biclustering approach to three real gene expression data sets and found the results very encouraging. In addition, we propose the average correlation value (ACV), a criterion to evaluate the property of a bicluster. This criterion has been compared with the mean squared residue score and ACV is found to be more appropriate.