Investigating a Breast Cancer Gene Expression Data Using a Novel Clustering Approach

Historically, breast cancer has been perceived as a disease with varying histological and clinical features. Breast cancer tumor classification is important in disease prognosis and prediction because different breast tumors respond differently to different treatments and have different survival rates. Gene expression profiling studies have increasingly been motivated in the past decades to develop a good classification of breast cancer in molecular subtypes, which can improve the standard clinical assessments by providing extra prognostic information. In this research, one of the most comprehensive breast cancer gene expression datasets is analyzed by applying a novel clustering approach to predict the breast cancer subtypes. The novel unsupervised clustering approach initially model the gene expression data as a network and employ a community detection method to identify network clusters. This method utilizes an efficient problem specific metaheuristic algorithm to optimize the modularity value and identify clusters of breast cancer samples with similar characteristics that presents different subtypes of breast cancer. To assess the significant of the newly defined breast cancer subtypes, we compared our findings with three breast cancer subtyping methods.