AI can reveal new cell biology just by looking at images
AI learns how to recognize and classify different dog breeds from images. A new machine learning method from CZ Biohub now makes it possible to classify and compare different human proteins from fluorescence microscopy images. Credit: CZ Biohub
Humans are good at looking at pictures and finding patterns or making comparisons. Take a look at a collection of dog photos, for example, and you can sort them by color, ear size, face shape, and so on. But can you compare them quantitatively? And perhaps more interestingly, can machines extract meaningful information from images that humans cannot?
Now Stanford University’s Chan Zuckerberg Biohub team of scientists have developed a machine learning method to quantitatively analyze and compare images—in this case microscopic images of proteins—without any prior knowledge. As reported in Natural Method, their algorithm, dubbed “cytoself”, provides rich and detailed information about the location and function of proteins within cells. This capability can speed up the research time of cell biologists and ultimately be used to speed up drug discovery and drug screening processes.
“It’s very exciting—we’re applying AI to new types of problems and still recovering everything that humans know, plus a lot more,” said Loic Royer, the study’s correspondent author. “In the future we can do this for many different types of images. It opens up a lot of possibilities.”
Cytoself has not only demonstrated the power of machine learning algorithms, it has also yielded insights into cells, the basic building blocks of life, and into proteins, the molecular building blocks of cells. Each cell contains about 10,000 different types of protein—some working alone, many working together, doing different jobs in different parts of the cell to keep it healthy. “A cell is much more spatially organized than we previously thought. That is an important biological result of how human cells are connected,” said Manuel Leonetti, also the study’s correspondent author.
And like all tools developed at CZ Biohub, cytoself is open source and accessible to everyone. “We hope this will inspire many people to use similar algorithms to solve their own image analysis problems,” Leonetti said.
Never mind a Ph.D., machines can learn on their own
Cytoself is an example of what is known as self-supervised learning, meaning that humans don’t teach any algorithms about protein images, as is the case in supervised learning. “In supervised learning, you have to teach machines one by one by example; it’s a lot of work and very tedious,” said Hirofumi Kobayashi, lead author of the study. And if machines are confined to categories that humans teach, it can introduce bias into the system.
“Manu [Leonetti] believe the information is already in the drawing,” said Kobayashi. “We wanted to see what the machine itself could do.”
Indeed, the team, which also includes CZ Biohub Software Engineer Keith Cheveralls, was surprised by how much information the algorithm was able to extract from the images.
“The level of detail in protein localization is much higher than we thought,” said Leonetti, whose group is developing tools and technology to understand cell architecture. “The machine converts each image of a protein into a mathematical vector. So you can start ranking images that look the same. We realized that by doing so, we could predict, with high specificity, the cooperating proteins in cells just by comparing the images, which was a bit surprising.” .”
First of its kind
While there has been some previous work on protein imagery using self-monitored or unsupervised models, never before has self-supervised learning been used with success on a dataset of over 1 million images spanning over 1,300 proteins measured from live human cells, said Kobayashi, an expert in machine learning and high-speed imaging.
The images are the product of the OpenCell CZ Biohub, a project led by Leonetti to create a complete map of human cells, including ultimately characterizing the 20,000 or so types of proteins that power our cells. Published earlier this year in Science was the first 1,310 proteins they tagged, including an image of each protein (produced using a fluorescent tag type) and a mapping of their interactions with each other.
Cytoself is a key achievement of OpenCell (all images available at opencell.czbiohub.org), providing very detailed and quantitative information on protein localization.
“The question of all the possible ways a protein can localize in a cell—all places and all kinds of combinations of places—is fundamental,” Royer said. “Biologists have been trying to establish all the possible places, for decades, and all the possible structures that exist in a cell. But that’s always been done by humans by looking at the data. The question is, how many limitations and biases do humans have to make This process isn’t perfect?”
Royer adds, “As we’ve shown, machines can do better than humans can. They can find better categories and see differences in very subtle images.”
The team’s next goal for cytoself is to track how small changes in protein localization can be used to recognize different cellular states, for example, normal cells versus cancer cells. This may hold the key to a better understanding of many diseases and facilitate drug discovery.
“Drug screening is basically trial and error,” says Kobayashi. “But with cytoself, it’s a big leap because you don’t have to do one-to-one experiments with thousands of proteins. It’s a low-cost method that can increase research speed significantly.”
AI program accurately predicts protein localization
Hirofumi Kobayashi et al, Self-supervised deep learning encoding high-resolution features of protein subcellular localization, Natural Method (2022). DOI: 10.1038/s41592-022-01541-z
Provided by Stanford University
Quote: AI can reveal new cell biology just by looking at images (2022, August 1) taken August 2, 2022 from https://phys.org/news/2022-08-ai-reveal-cell-biology-images.html
This document is subject to copyright. Apart from any fair dealing for purposes of personal study or research, no part may be reproduced without written permission. Content is provided for informational purposes only.
#reveal #cell #biology #images
Comments
Post a Comment