Community detection from genomic datasets across human cancers

Cancers originating from different organs can show similar genomic alterations whereas cancers originating from the same organ can vary across patients. Therefore cancer stratification that does not depend on the tissue of origin can play an important role to better understand cancers having similar genomic patterns irrespective of their origins. In this work, we formulated the problem as a weighted graph and communities were found using a modularity maximization based graph clustering method. We classified 3,199 subjects from twelve different cancer types into five clusters. The five communities show significantly different survival rate curves. The distribution of tumor types against communities shows that lung, colon and rectum adenocarcinoma cluster together, whereas breast and ovarian cancers form another cluster.