Journal of Health and Biomedical Informatics

Introduction: The microarray technology and production of gene expression data are among the important developments in genetic science that provide ability to study the behavior of thousands of genes, simultaneously.  Clustering is one of the most important data mining techniques used in gene expression data analysis. As, the performance of clustering methods is strongly affected by the structure of data, the result of clustering is always uncertain and there is no algorithm that can be used for all kinds of data. In this study, ensemble clustering (combined results of multiple clustering algorithms) was used for gene expression data analysis rather than using a single algorithm.

Methods: The performance of ensemble clustering in three gene expression data sets, Nutt-v3, Alizadeh-v2 and SU, were evaluated by adjusted Rand index. Twelve different clusterings resulted from the combination of four clustering algorithms with three dissimilarity matrices were simultaneously applied on data. After merging the results,and running the final clustering, the estimated clusters were compared with actual groups by the adjusted Rand index.

Results: The adjusted Rand index for the three data sets of Nutt-v3, Alizadeh-v2 and SU, were respectively 1, 0.9 and 0.58 which shows the remarkable accuracy of the proposed method in detecting patterns in data sets. Moreover, the designed algorithm could detect the actual number of clusters without errors.

Conclusion: Ensemble clustering is a powerful and reliable method for gene expression data analysis. Due to the accuracy and quality of this method in detection of real data structures, it can be replaced the individual clustering algorithms.