Combining Multiple Clustering and Network Analysis for Discoveries in Gene Expression Data

Clustering is a challenging research task which could benefit a wide range of practical applications, including bioinformatics. It targets success by optimizing a number of objectives, a characteristic mostly ignored by clustering approaches. This paper describes a synthetic clustering algorithm which first applies multi-objective based approach to produce the alternative clustering solutions. Then the best clusters from each solution are selected and combined into a seed for a compact and effective solution which is expected to be better than all the individual solutions because it combines the best of each. This way, the developed algorithm may be classified as a fuzzy clustering approach because each object may belong to more than one cluster in the synthesized solution with a degree of membership in each cluster. Another interesting aspect of the algorithm is that it identifies the outliers. Further, a network is built from the relationships of the objects within the various clusters. The network is analyzed to reveal interesting discoveries not clearly reflected in the clustering outcome. The validity and applicability of the presented methodology has been assessed using synthetic and real data from the cancer.