Biclustering Analysis Using Plaid Model on Gene Expression Data of Colon Cancer

Unlike other typical clustering analysis, which considers column only, biclustering analysis processes a matrix into sub-matrices based on rows and columns simultaneously. One method of bicluster analysis uses the probabilistic model, like the plaid model, that provides overlapping bicluster. The plaid model calculates the value of an element given from a particular sub-matrix for each cell; thus, the value can be seen as the number of contributions of a particular bicluster. The algorithm begins with preparing the input data as a matrix, then an initial model is assessed and makes a residual matrix from the model. After that, we determine bicluster candidates, which are evaluated for its effect parameters and bicluster membership parameters. Finally, the bicluster candidate is pruned to give the optimal bicluster. We implemented the algorithm on gene expression dataset of colon cancer, where the rows and columns contain observations and types of genes, respectively. We carried out in six distinct scenarios in which each scenario uses different model parameters and threshold values. We measured the results using Jaccard index and coherence variance. Our experiments show that biclustering analysis on a model with mean, row, and column effects of colon cancer data output low coherence variance.