Clustering High-Dimensional Data: The Expression of E-cadherin, CD44 and p53 Molecules in Lip Cancer

Objective: Clustering techniques can determine which expression patterns are important and which genes contribute to such patterns. We evaluate performance on data from a lip carcinoma study in Greece. Lip carcinoma is one of the most common malignant oral and maxillofacial tumours and in advanced clinical stages has a poor prognosis. E-cadherin, CD44 and p53 molecules are associated with cellular adhesion. Material and Methods: To prepare for clustering, we divided each of the median normalized gene expression values by the range of that gene. Next, we set our prior parameters and we performed the final inference using pooled sets of Markov chain Monte Carlo (MCMC) runs. After pooling the chains, we grouped the data into clusters and selected E-cadherin, CD44 and p53 molecules using the marginal median model as cut off. The selection of a small set of genes is advantageous here. A small number of selected genes is appealing to biologists because they constitute a manageable set of candidates on which further studies can be performed. Results: E-cadherin, CD44 and p53 molecules were selected as discriminatory. Results highlight the fact that clustering method has successfully selected genes that are biologically consistent with current research and that provide strong biological validation of the cluster configuration suggested. Conclusion: A clustering method that takes advantage of known substructure in the data when simultaneously clustering high-dimensional data with an unknown number of clusters, and selecting the best discriminating variables for those clusters implies the opportunity to handle bigger datasets. When analyzing real data, clustering has found three genes that agree with current biological research and literature and that provide biological validation of the cluster configuration. Overall, clustering can provide biologists with both useful and manageable information for further experimental research.