Biclustering Using Modified Matrix Bandwidth Minimization and Biogeography-based Optimization

TL;DR

This paper introduces a novel biclustering algorithm that uses a modified matrix bandwidth minimization approach combined with biogeography-based optimization to identify clusters in two-mode data without prior knowledge of cluster count.

Contribution

It adapts the bandwidth minimization algorithm for non-square, non-binary matrices and integrates biogeography-based optimization for improved biclustering performance.

Findings

Successfully reveals underlying biclusters in data

Operates without prior knowledge of number of biclusters

Shows potential for further research in two-mode data analysis

Abstract

Data matrix having different sets of entities in its rows and columns are known as two mode data or affiliation data. Many practical problems require to find relationships between the two modes by simultaneously clustering the rows and columns, a problem commonly known as biclustering. We propose a novel biclustering algorithm by using matrix reordering approach introduced by Cuthill-McKee's bandwidth minimization algorithm, and adapting it to operate on non-square and non-binary matrices, without the need to know apriori the number of naturally occurring biclusters. This transforms a two-mode matrix into almost block diagonals, where the blocks indicate the clusters between the two modes of the matrix. To optimize the bandwidth minimization problem, we adapted the Biogeography-based Optimization algorithm using logistic equation to model its migration rates. Preliminary studies indicate that this technique can reveal the underlying biclusters in the data and has potential of further research for two-mode data analysis.