Engineer creates unique software that predicts stem cell fate

The software, developed by Andrew Cohen, an assistant professor of electrical engineering, analyzes time-lapse images capturing live stem cell behaviors. It will allow scientists to search for mechanisms that control stem cell specialization, the main obstacle in advancing the use of stem cell therapy for treatment of disease. It could also lead to new research into causes of cancer, which involves cells that continuously self-renew.

Stem cells play a key role in human development, and also offer the potential to repair tissues or organs damaged by disease or injury. But, in order to use stem cell-based therapies, biologists need to better understand the mechanisms that control stem cell differentiation.

"This is a brand-new set of tools for developmental biologists," says Cohen, "and it supports an area where no other predictive solutions exist."

The research is published Feb. 7 in the journal Nature Methods. Co-authors are Michel Cayouette and Francisco Gomez neurobiologists at the Institut de recherches cliniques de Montreal, and Badri Roysam, a computer engineering professor at Rensselaer Polytechnic Institute.

The software is 87 percent accurate in determining the specific "offspring" a stem cell will ultimately produce, and 99 percent accurate in predicting when self-renewal of these stem cells will end in specialization.

A hunt for markers

As an example of the software's utility, Cohen cites using stem cells to treat the eye disease macular degeneration. The stem cells would need to produce more photoreceptor neurons for treatment to succeed. "But if you simply implant the stem cells into the retina, there are other types of cells that could develop," he says, "and that could potentially make the patient's vision worse."

Finding a solution has been hampered by the fact that there are very few markers that can predict cell division outcomes.

Subtle behaviors that characterize populations of stem cells with different fates are difficult or impossible for human observers to recognize. Cohen's tool, which runs on a standard PC, is able to track and generate predictions for up to 40 cells in real time. It outperforms the human eye in detecting differences in how the cells change over time.

Current methods of observing live cells produce terabytes of data, a volume that requires massive amounts of computation to find the most relevant information. A new computer cluster in CEAS was acquired for just this kind of research. To manage the predictive aspects of the program, Cohen used a uniquely sensitive mathematical approach based on algorithmic information theory.

Answers in DNA

Scientists know little about programming of stem cell outcomes except that it is a multifaceted process.

"In many cases, stem cells take their developmental cues from their environment," says Cohen. "Part of the programming mechanism is determined by surrounding cells. But once these cells begin to develop in a particular way, their offspring continue down that path even if the environment changes. So at some point they have been programmed to their fate."

The researchers designed the software to be used for isolating the genes and proteins that control the specialization process, which could allow researchers to identify and ultimately manipulate these programmed mechanisms.

Brian Link is a developmental biologist at the Medical College of Wisconsin who works with Cohen but is not an author on the Nature Methods paper. The two will be putting the software to the test to study behaviors of organelles within the cell as indicators of stem cell fate.

"The method isn't perfect," says Link. "It doesn't tell us about the influence of the behaviors. It tells us that a particular behavior is important, but it doesn't tell us how."

Still, the tool has already proven itself, he says. In a study of stem cells from the retinas of rats, Cohen's software independently confirmed the significance of at least one of the cell behaviors that Link's lab had previously identified using a gene manipulation technique.