COMPARISON OF BICLUSTERING ALGORITHMS FOR DETECTION OF NOISY AND OVERLAPPING BICLUSTERS USING SIMULATED GENE EXPRESSION DATA

Biclustering is an important technique for pattern recognition in gene expression data to find groups with similar expression patterns. Issues exist with biclustering algorithms in general and it is not clear which algorithms are best suited for this task. The present study evaluated four biclustering algorithms using simulated data for efficacy of detection of overlaps. Scenarios were constructed by changing the size of the data matrix and the level of noise and overlap. Results showed that the Cheng and Church and Spectral algorithms were not sufficient for these scenarios. The BiMax algorithm was robust to noise but its efficacy decreased in the presence of overlap between biclusters. The Plaid algorithm was mostly robust for overlap, but its efficacy decreased as the noise level increased. These results are designed to aid researchers when selecting the most appropriate algorithm for a dataset.