Multi-Level Data Analysis in Cancer: Tools and Approaches

This chapter reviews the algorithmic principles underlying the majority of existing approaches for deconvoluting transcriptomic tumoural profiles and considers the major difficulties met in the application of those algorithms in practice. It highlights that a family of methods for unsupervised deconvolution remains under-explored in the field. The chapter suggests that this approach, properly used, can lead to the improved definition of context-specific molecular profiles of cell types present in the tumoural microenvironment. Biology is organized hierarchically, from lower levels that comprise DNA, RNA and proteins, over cells, to multicellular organisms and ecosystems. Cancer is characterized by a distinct set of biological capabilities, for example resistance to cell death, or the ability to sustain proliferative signalling. One of the challenges tackled by the computational systems biology of cancer consists of determining the cellular composition of a bulk tumoural sample, where cells of different types, including tumoural cells, are mixed together and collectively contribute to the measured molecular profiles.