Tom Kirchhausen’s work is characterized by use of emerging technologies — from the early days of molecular cloning to contemporary high-resolution structural visualization and live-cell imaging. He used the tools of x-ray crystallography, cryo electron microscopy, and single-molecule biophysics to create a “molecular movie” of clathrin-mediated endocytosis, and in this way relate these molecular events to functional properties of the surfaces of living cells.
He also uses frontier optical-imaging modalities to examine these, and other cellular membrane remodeling processes exemplified by cell size regulation and organization of the ER during cell division, post mitotic formation of nuclear pores complexes, the generation of intraluminal vesicles in endosomes, and the dynamics of virus-host cell interactions during infection. The richness and magnitude of the data acquired over periods ranging from seconds to hours, creates new challenges for obtaining quantitative representations of the observed dynamics and for deriving accurate and comprehensive models for the underlying developmental mechanisms.
Combined with more recent use of deep learning methods to help analyze these complex data sets, he aims to generate other ‘molecular movies’ as new frameworks for analyzing some of the molecular contacts and switches that participate in the regulation, availability, and intracellular traffic of the many molecules involved in signal transduction, cell host – virus interactions, immune responsiveness, lipid homeostasis, cell-cell recognition and organelle biogenesis of importance for the understanding of many diseases including cancer, viral infection and pathogen invasion, Alzheimer’s, as well as other neurological diseases.
About his talk: Revealing Biology with Deep Learning and Frontier Microscopy
Today, we are witnessing an artificial intelligence inspired transforming revolution that is helping set up new visualization standards for frontier optical and electron imaging modalities to analyze and understand sub-cellular processes from single molecules to entire organisms in the complex and dynamic three-dimensional environment of livingcells in isolation and within tissues.
This talk will illustrate how we make use of deep learning artificial intelligence (AI) algorithms to explore temporal time resolved 3D image sets of cell biological processes acquired using lattice light-sheet microscopy (LLSM) and high-resolution volumetric snapshots of cells obtained using focused ion beam scanning electron microscopy (FIB SEM).
In one example, temporal 3D experiments previously limited to seconds or minutes by photo-bleaching or by photo-toxicity, can now be done with extremely low photon doses at diffraction limited resolution and high-temporal precision with unprecedented duration of minutes or hours. We will describe how use of deep learning-aided image processing of LLSM data following SARS-CoV-2 and AAV infections or endocytosed oligonucleotides and proteins helped us reveal unexpected entry pathways leading to successful viral entry and delivery of biologicals into the cytosol.
As a second example, we developed an automated image segmentation tool able to extract data from cells volumetrically imaged using FIB SEM. We trained neural networks to identify mitochondria, ER and Golgi apparatus and subcellular structures like nuclear pores, clathrin coated pits and clathrin coated vesicles in complete cells in fractions of hrs rather than the many hours or weeks of laborious manual work. We illustrate how using these tools help us learn about new topological features of the mitochondria and ER and about the organization of nuclear pore complexes.