Using cutting-edge super-resolution microscopy technology, expansion microscopy, we discovered that tubular nuclear envelope invaginations are highly abundant in vertebrate embryonic cells. These structures are poised to extend the role of the nuclear envelope in regulating gene expression deep into the nucleus. Shedding light on this phenomenon requires segmenting the 3D structure of invaginations in a huge dataset of microscopy data. We are interested in utilizing learning algorithms and in particular deep neural networks for this 3D segmentation task.
At key points within neural circuits, neurons integrate information from multiple sources to make a choice. We are interested in unraveling how such choices are implemented by the circuits, by developing generative probabilistic models of neural activity in multiple neural populations involved in making a decision and comparing these predictions to experimental measurements of neural activity. In particular, we will focus on the circuit mediating the choice of the response type a larval zebrafish would present in the face of an alarming stimulus.
