Hierarchical Clustering of Microarray Data with Probe-level Uncertainty

Handling microarray data is particularly challenging mainly due to the high dimensionality of such data, which demands for computer-aided methods, and to the intrinsic difficulty of devising notions of proximity between spots of array traps. In this paper, we propose a new approach to modeling the probe-level uncertainty in microarray data that allows for a more expressive representation of the data and a more accurate processing. This approach is essentially based on a recently proposed method for uncertain data clustering. This method lies in a centroid-linkage-based agglomerative hierarchical algorithm, named U-AHC, and an information-theoretic-based distance measure between uncertain data. We have conducted experiments on four large microarray datasets, in order to assess effectiveness of the proposed clustering method. Experimental results have shown high quality results in terms of compactness of the clustering solutions.