A Weakly Supervised Clustering Method for Cancer Subgroup Identification

Identifying subgroups of cancer patients is important as it opens up possibilities for targeted therapeutics. A widely applied approach to group with unsupervised clustering techniques based on molecular data tumor samples. The patient clusters are found be interest if they can associated a clinical outcome variable such the survival patients. However, these variables do not participate in decisions. We propose an approach, WSURFC (Weakly Supervised Random Forest Clustering), where process weakly supervised interest. supervision step handled by learning similarity metric features that selected predict this variable. More specifically, involves random forest classifier-training variable, case, class. Subsequently, internal nodes used derive among pairs In way, utilizes nonlinear subspace original learned classification step. first demonstrate hand-written digit datasets, able capture salient structural similarities pairs. Next, we apply find breast subtypes using mRNA, protein, and microRNA expressions features. Our results show could identify interesting more effectively than adopted methods.