An emerging approach in systems biology and personalized medicine is that of relating molecular networks to disease outcomes and treatment. In a nutshell, the idea is that networks that describe biology relevant to disease phenotypes may differ between patients, or between patient subpopulations, such that their systematic characterization could help to explain corresponding variation in disease phenotypes or response to therapy. A major computational and experimental challenge is to develop algorithms and protocols by which to learn such networks. Using protein signalling as an paradigmatic example, we will discuss our ongoing efforts to develop approaches for network inference in this setting, including scalable tools for time-course data, joint estimation of multiple (related) networks, non-linear models and validation frameworks that can be applied in the complex mammalian settings relevant to disease biology. Along the way we will discuss some of the caveats and fundamental concerns in the general area of causal networks for biological applications.