Advances in observation instruments and abundance of computational power for simulations encourage scientists to gather and produce unprecedented amounts of increasingly complex data. Organizing data automatically to enable efficient and unobstructed access is pivotal for the scientists. Organizing these vast amounts of complex data, however, is particularly difficult for scientists who have little experience in data management; hence they spend considerable amounts of time dealing with data analysis and computing problems rather than answering scientific questions or developing new hypotheses. Therefore scientific experiments are in many ways ideal targets for research in self-managing database systems. In this paper, we describe challenges and opportunities for research in automating scientific data management. We first discuss the problems faced in particular scientific domains using concrete examples of large-scale applications from neuroscience and high-energy physics. As we will show, the scientific questions are evolving ever more rapidly while datasets size and complexity increases. Scientists struggle to organize \& reorganize the data whenever their hypothesis change and therefore their queries & their data changes as well. We identify research challenges in large-scale scientific data management related to self-management. By addressing these research challenges we can relieve the burden of organizing the data off the scientists, thereby ensuring that they can access it in the most efficient way and ultimately enabling the scientists to focus on their science.