A New Approach for Clustering Gene Expression Profiles

Microarray experiments generate a considerable amount of data. Analyzing those data properly would help us gain a huge amount of biologically relevant information about the global cellular behavior. Clustering is one of the first steps in data analysis of high-throughput expression measurements. Many clustering algorithms have proved useful to make sense of such data. These algorithms, though useful, suffer from several drawbacks. Here, we propose an iterative two-step clustering algorithm which tackles some of these drawbacks. In the first step, a new graph-theoretic approach is introduced to locate clusters. In the second step, the radius(or size) of each cluster is identified adaptively. Our method doesn’t need to predefine the cluster number or cut the tree structure as K-means or hierarchical clustering does. The algorithm is successfully validated using existing data sets and can outperform hierarchical and K-means clustering in some aspects.