Integrated Clustering of Cancer Genes Based on Local Weighting

This paper proposes propose a novel ensemble clustering method to address the problem that integrated clustering typically treats all component clusters equally without considering their reliability, thereby being vulnerable to low-quality component clusters. Firstly, five clustering methods are adopted to generate different component clusters. Secondly, a new ensemble-driven clustering index (ECI) based on the unreliability of component clusters is introduced to construct a pool of candidate component clusters and discard low-quality ones. To capture the local diversity of the ensemble, we utilize a matrix known as a locally weighted co-association matrix (LWCA), and a new consistency function, improved locally weighted graph partitioning with consistency consideration (RLWGP), we propose a further approach by incorporating cluster labels across the ensemble through an entropy criterion. In this approach, we consider both clusters and objects as nodes in the graph, and the advantage of bipartite graph structure facilitates efficient graph partitioning and preferable clustering performance. Our experimental results, conducted on various cancer datasets, demonstrate that the proposed method outperforms existing methods in terms of accuracy, precision, and recall.