Choosing the right treatment for a given patient is a critical task in cancer care. Suboptimal treatment choices have a number of negative consequences including extreme financial waste, devastating treatment side effects, and the potential for tumour progression. Treatment selection is often guided by the presence of biomarkers, molecular measurements derived from a patient's tumour that can indicate sensitivity or resistance to a given drug. For a number of reasons biomarker discovery research is done on pre-clinical models, chief among them are cell line viability screens. In this talk I will discuss the challenges of finding biomarkers in cell line screens, with a focus on the kinds of noise that plague these datasets. Then I will discuss how tools from computational topology are ideally suited for this noisy environment. In particular I will discuss how building a discrete simplicial complex with drugs as the 0 simplices and using the HodgeRank procedure can find interpretable and self-certifying orderings on drugs and how this procedure can be personalized for drug recommendation.