Using complex networks for refining survival prognosis in prostate cancer patient

Complex network theory has been used, during the last decade, to understand the structures behind complex biological problems, yielding new knowledge in a large number of situations. Nevertheless, such knowledge has remained mostly qualitative. In this contribution, I show how information extracted from a network representation can be used in a quantitative way, to improve the score of a classification task. As a test bed, I consider a dataset corresponding to patients suffering from prostate cancer, and the task of successfully prognosing their survival. When information from a complex network representation is added on top of a simple classification model, the error is reduced from 27.9% to 23.8%. This confirms that network theory can be used to synthesize information that may not readily be accessible by standard data mining algorithms.