In 2014, Hirofumi Kobayashi was getting his master’s degree when the world of microscopy opened up to him. The Nobel Prize in chemistry had just gone to a trio of researchers who developed a way to explore biology at the nanoscale, using fluorescence to visualize molecular action within cells. “That was the trigger,” said the biochemist, who at the time was tracking viruses at the University of Tokyo. He decided to dedicate his work to using optics to probe biology’s secrets.
Not only could biological images reveal a world of undiscovered cell dynamics, Kobayashi thought, but they were simply beautiful to look at. Later, while getting his Ph.D., Kobayashi developed his own instrument to visualize cancer cells in blood. He soon turned to artificial intelligence to decipher their contents, developing an algorithm that recognized minute changes in the shape of cancer cells — undetectable to most humans — to determine whether they were responding to drug therapy.
Today, Kobayashi continues to turn the beautiful into something usable at the Chan Zuckerberg Biohub, combining his optics training with new computational tools. Drawing on images from a trove of more than 1,300 cell images called OpenCell, he trained a deep learning model to automatically develop an atlas of where proteins reside on a subcellular level. “On one side we can save the human labor, on the other side, we can remove or diminish the human bias,” said Kobayashi. “And — another extreme — we hope that maybe we can find something that human beings cannot.” Sometimes in science, the beauty is in what you can’t see.
— Katie Palmer
