• Feature
  • Genetic Networks

Algorithm holds potential to change cell biology

by Eva Voinigescu May 17 / 17
A new deep neural network algorithm is promising to change the way cell biologists work.

The algorithm automatically analyzes data from microscopes, scrutinizing visual patterns in the cell to determine how proteins function and what goes wrong in disease. The research was supervised by CIFAR fellows Brenda Andrews, Charles Boone and Brendan Frey and was the cover story in the latest issue of Molecular Systems Biology. All three researchers are members of CIFAR’s Genetic Networks program. Frey is also a member of the Learning in Machines & Brains program.

In a field where manual image analysis is still standard practice, automated microscopes are capturing data much faster than it can be analyzed. This new system, called DeepLoc, could save scientists weeks or months of analysis all while providing more accurate results.

“We can learn so much by looking at images of cells: how does the protein look under normal conditions, and do they look different in cells that carry genetic mutations or when we expose cells to drugs? People have tried to manually assess what’s going on with their data but that takes a lot of time,” says Ben Grys, a graduate student at the University of Toronto and a co-author of the paper.

Algorithm holds potential to change cell biology

DeepLoc takes micrographs of yeast cell proteins like the ones seen above and automatically classifies them, allowing researchers to analyze protein functions and determine what goes wrong in disease, much faster and with greater accuracy. 

DeepLoc greatly improves on earlier machine learning techniques that have been used for this type of microscopic image analysis but are slower and not easily trained on new or different sets of data.

The key to DeepLoc’s success is a deep learning method that allows the algorithm to learn by itself directly from pixel data taken from images of cells, rather than needing detailed instructions, which was necessary with previous computer vision systems.

This allows DeepLoc to quickly classify data sets from sources other than the one it was trained on. Researchers including Grys, Oren Kraus, Jimmy Ba and Yolanda Chong trained the algorithm on a comprehensive dataset of 4,000 yeast proteins from research previously done by Grys and Kraus. The set had previously been analyzed in 2015 using an older computer vision method that took months to complete. DeepLoc analyzed the same data in hours and detected changes that the old technology couldn’t, like that resulting from hormone treatment.

In order to classify new data sets that are significantly different from the ones it was trained on, DeepLoc uses a method called transfer learning, in which knowledge gained from solving one problem is stored and then applied to a different but related problem.

“Someone with some coding experience could implement our method. All they would have to do is feed in the image training set that we’ve provided and supplement this with their own data. It takes only an hour or less to retrain DeepLoc and then begin your analysis,” says Grys.

Grys and Kraus were able to retrain DeepLoc with markedly different images from another lab within hours. The researchers say DeepLoc’s application by the wider cell biology community could hasten discoveries about key protein function, and Kraus and Ba are already working to commercialize the method, with the hope of applying it to pharmaceutical development.

“We hope to make the early drug discovery process all the more accurate by finding more subtle effects of chemical compounds,” says Kraus.