5/31, LSBM Monday Seminar Series #8, Hirofumi Kobayashi, Self-supervised learning resolves subcellular protein localization and more

Presenter: Hirofumi Kobayashi (Chan Zuckerberg Biohub)
Date: May 31st (Mon), 12pm-

 

 

About presenter: 

Hirofumi is a recipient of the Japan Society for the Promotion of Science overseas fellowship. Currently, his focus is on developing novel techniques for image restoration and image analysis using deep learning techniques. Hirofumi majored in applied chemistry, where his thesis topic was generating interstellar molecules. During his Master's degree, he developed a replication-incompetent bivalent vaccine against influenza and parainfluenza viruses. After 1 year of research into regenerative medicine using iPS cells, and cell metabolism, Hirofumi pursued his Ph.D. developing ultra high-speed imaging flow cytometry and image analysis pipelines. Using deep learning, he analyzed drug susceptibility of leukemia cells in blood without dilution or hemolysis.

 

Abstract:

Elucidating the wiring diagram of the human cell is one of the central goals of the post-genomic era. For this purpose, I and my colleagues developed cytoself[1], a deep learning-based approach for fully self-supervised protein localization profiling and clustering. cytoselfleverages a self-supervised training scheme that does not require pre-existing knowledge, categories, or annotations. To demonstrate the utility of cytoself, we applied cell images from OpenCell [2], an interactive database that consists of confocal images and mass spectrometry data for a library of 1,311 CRISPR-edited cell lines harboring fluorescent tags to systematically map protein localization in live cells and protein interactions under endogenous expression conditions. We found that the representations derived from cytoselfencapsulate highly specific features that can be used to derive functional insights for proteins on the sole basis of their localization. I will present the working mechanism as well as how cytoselfcan resolve subcellular protein localization from multiple aspects. I will also discuss how the representations of cell images could predict molecular interactions.

[1] Kobayashi, Hirofumi, et al. "Self-Supervised Deep-Learning Encodes High-Resolution Features of Protein Subcellular Localization." bioRxiv(2021).
[2] Cho, Nathan H., et al. "OpenCell: proteome-scale endogenous tagging enables the cartography of human cellular organization." bioRxiv(2021).