Despite huge research advances, cancer is still often a deadly disease. A general trend in pre-clinical research is to collect a lot of molecular data about the tumors, which is now becoming easier thanks to relatively fast and affordable technologies. Detecting from these data whether and which molecular, regulatory mechanisms within the cancerous cells are faulty is challenging, but it poses great opportunities for therapies. We are using network-based approaches, meaning that we focus on how genes, proteins and other molecules of the cell, such as metabolites, are connected. The links in the networks represent interactions between molecules, and thus comparing links of networks derived from different patient groups tells us about differences in their molecular interactions. These can be translated to drug action: Drugs may be able to reinstall the correct, healthy molecular interaction pattern within cells, or might be differently affected by them. Thus, the information on altered network links can be exploited to predict which drugs or drug combinations may be most beneficial to a certain patient group. An important step in my analyses is to use the structural properties in the networks to reduce errors introduced by the underlying measurements. This correction step is key to robust and reproducible predictions on altered regulations and thus drug action, and it paves the way to patient-tailored therapies.