The objective of the proposed research is to find interpretable dynamical features from neuroimaging data and fuse them with structural features in the context of mental disorders. The brain is a highly dynamical system, and mental disorders are often thought to partially stem from disordered brain dynamics. However, at the whole-brain level, approaches that focus on extracting dynamical features in a data-driven and psychiatrically interpretable way are understudied. Moreover, work that links the structure of the brain to its function, is limited in its empirical predictive power, especially in capturing important subject-specific dynamical features. It is thus the goal of this research to both develop machine learning methods that can extract dynamical features from empirical timeseries, and link structure to function in a flexible, and data-driven manner.