Biomedical Application of Fuzzy Association Rules for Identifying Breast Cancer Biomarkers

Current breast cancer research involves the study of many different prognosis factors: primary tumor size, lymph node status, tumor grade, tumor receptor status, p53, and ki67 levels, among others. High-throughput microarray technologies are allowing to better understand and identify prognostic factors in breast cancer. But the massive amounts of data derived from these technologies require the use of efficient computational techniques to unveil new and relevant biomedical knowledge. Furthermore, integrative tools are needed that effectively combine heterogeneous types of biomedical data, such as prognosis factors and expression data. The objective of this study was to integrate information from the main prognostic factors in breast cancer with whole-genome microarray data to identify potential associations among them. We propose the application of a data mining approach, called fuzzy association rule mining, to automatically unveil these associations. This paper describes the proposed methodology and illustrates how it can be applied to different breast cancer datasets. The obtained results support known associations involving the number of copies of chromosome-17, HER2 amplification, or the expression level of estrogen and progesterone receptors in breast cancer patients. They also confirm the correspondence between the HER2 status predicted by different testing methodologies (immunohistochemistry and fluorescence in situ hybridization). In addition, other interesting rules involving CDC6, SOX11, and EFEMP1 genes are identified, although further detailed studies are needed to statistically confirm these findings. As part of this study, a web platform implementing the fuzzy association rule mining approach has been made freely available at: http://www.genome2.ugr.es/biofar .