Mental health conditions like depression are most often studied cross-sectionally, using methods that are blind to the dynamic properties of symptoms, cognition, and physiology. Complex systems theory has emerged as a popular framework for capturing this, offering a theoretical path to understanding the onset and resolution of depression. Although there is much excitement about these ideas (including network theory of mental disorders), data has been less forthcoming – in part owing to practical challenges in acquiring high quality, repeated within-person assessments from depressed individuals, or those at-risk. In this talk I will describe some recent efforts to remedy this, using a variety of novel data sources, that vary in depth, breadth (and validity). These include routine clinical data from UK NHS patients, language in social media posts, passively gathered measures of processing speed and experience sampling data from a smartphone app. I will use these data to illustrate important confounds inherent in cross-sectional approaches, how novel data sources can allow us to construct intra-individual time series data. I will share some key empirical observations– including the finding that personalised network connectivity of symptoms is associated with more extreme fluctuations in depression over time, and evidence that depression causes certain cognitive impairments rather than the other way around.