Computer Vision Interest Group Seminar: “Neurotransmitter Classification from Electron Microscopy” (HHMI Janelia)

Speaker:  Jan Funke, Group Leader at HHMI Janelia

Topic:  Neurotransmitter Classification from Electron Microscopy

In this talk, we will have a look at our ongoing work to discover signals in microscopy images that we did not know were in there. Specifically, we will show that we can predict the neurotransmitter identity of Drosophila neurons
from electron microscopy images of synapses using a simple deep neural network. Effectively, this allows us to predict properties of function from
ultra-structure alone, which is somewhat surprising since humans do not seem to
be able to do the same. We will go on to show how, once trained, the same deep
neural network can then be used to “explain” to us how it made its decisions
and thereby guide us in furthering our understanding of the relation between
function and structure.

Parts of this work have been published as preprints:
https://www.biorxiv.org/content/10.1101/2020.06.12.148775v2
https://arxiv.org/abs/2109.13412v1