Using Computer Visualization to Predict Stem Cell Behavior

Computer vision can predict what will happen to stem cells once they divide.

SHARE

By Marlene Cimons, National Science Foundation

Scientists for the first time have used computer vision technology to predict what will happen to stem cells once they divide, that is, whether they will continue to produce additional stem cells, or become part of a future developing organ.

The advance, developed by researchers at Rensselaer Polytechnic Institute, could lead to more effective methods of growing stem cells on a large scale basis for therapeutic use.

“Achieving large-scale production of stem cells has been very difficult; you need almost an industrial size process,” said Badri Roysam, professor of electrical, computer and systems engineering at Rensselaer, one of the scientists involved in the project. “If we can have a computer watch over every cell and basically sort them into different categories, it would give you the basis for large scale stem cell production.”

Stem cells do one of two things when they divide. So-called self-renewing cells split into two daughter cells, at least one of which is a stem cell that can make additional stem cells. The other, known as a terminally differentiated cell, divides only once and the resulting cells are specialized, meaning they are no longer stem cells. For the latter, “it’s the end of the line as a stem cell,” Roysam said.  “They no longer have their ‘stem-ness.”  

In order to develop effective stem cell therapies, researchers need access to large numbers of specific cells, which is difficult because there are no existing ways to control or manipulate the division of bulk quantities of cells. If they knew in advance which way the cells would go after they divide, researchers likely could influence the division process in order to produce large numbers of the correct type of cells. 

The new computer vision system takes images of cells every five minutes, and uses a process called algorithmic information theoretic prediction (AITP) to watch the behavior of the cells, analyze the behavior and decide whether each individual cell will split into self-replicating or terminal daughter cells. The process happens in real time; thus, the researchers know what will happen to the cells before they actually divide.

“It’s like a movie, a time-lapse microscope movie,” Roysam said. “It’s like what they do on a movie screen with animation.” 

Roysam and his former student, Andrew Cohen, now assistant professor of electrical engineering and computer science at the University of Wisconsin, Milwaukee, used rat retinal progenitor cells that were cultured in a collaborator’s laboratory at McGill University. They were able to predict with 99 percent accuracy whether the rat retinal cells would split into self-renewing or specialized cells.

Moreover, in the case of specialized cells, they could predict with 87 percent accuracy the type of specialized cell it would become.  “Within the retina, there are many different types of cells,” Roysam said. “We could tell with 87 percent accuracy exactly what subtype of cell it would become.”

The results suggest that stem cells “display subtle dynamic patterns that can be sensed computationally to predict the outcome of their next division,” he added. “In theory, AITP can be used to analyze nearly any type of cell, and could lead to advances in many different fields.”

The method, in fact, could “be beneficial for one day taking cells from a patient, and then growing large amounts of the kind of cells that patient is in need of,” he added. “This could enable many new and exciting types of medical treatments using stem cells.”

Roysam said that the impact on the field of cell biology could be profound. “Cells are giving out hints of what they are about to do, and we cannot decode these hints when we are viewing them just by eye. Now you have a computational eye that can see subtle things the human eye cannot see. As a computer scientist, I am hugely excited about the future role of computers. Computational sensing is going to become very big in the future of biology.”

Roysam hopes, among other things, to test computational sensing on human stem cells. “There is an exponential manner in which our state of knowledge of stem cells is growing,” he said. “As a scientist, I want to see the fundamental mechanisms that underlie all these things unravel to the fullest. The most exciting aspect of this work is how people across disciplines can collaborate--that we’ve found ways to combine our knowledge to make an advance. I think we should see a lot more of that.”