Method for Joint Clustering in Graph and Correlation Spaces

Abstract: Graph algorithms are often used to analyze and interpret biological data. One of the widely used approaches is to solve the problem of identifying an active module, where a connected subgraph of a biological network is selected, which best reflects the difference between two biological states being considered. In this work, we extend this approach to the case of a larger number of biological states and formulate the problem of joint clustering in graph and correlation spaces. To solve this problem, an iterative method is proposed, which takes as the input the graph G and the matrix X, in which the rows correspond to vertices of the graph. As the output, the algorithm generates a set of subgraphs of graph G so that each subgraph is connected and the rows corresponding to its vertices have a high pairwise correlation. The efficiency of the method is confirmed by an experimental study using simulated data.