This project aims at using data-driven methods, in particular based on deep learning, to study the dynamics of the coherent structures in turbulent flows. We will focus on intense Reynolds-stress events and on vortical structures, and we will consider canonical wall-bounded turbulent flows. These structures significantly contribute to the momentum transfer and the dissipation in wall-bounded turbulence, which are essential mechanisms inherent to the dynamics of the flow. We will identify these structures, track them in time, and use deep-learning methods (transformers and graphnets) to reproduce the evolution in time of these structures. Doing so, we will obtain significant insight into the details of these interactions in turbulence, which is an important step towards developing robust strategies for flow control.