Paving the Way to Alzheimer’s Cure, One Algorithm at a Time

When chemistry Nobel laureate Michael Levitt met his wife two years ago, he didn’t know it would lead to a wonderful friendship with a young Israeli scientist. When Israeli scientist Shahar Barbash decided to found a startup with the aim of cutting down the time needed to develop new medicine, he didn’t know that a friend’s wedding will help him score a meeting with a man many want to meet but few do. But Levitt’s wife is an old friend of Barbash’s parents, and the rest, as they say, is history.

For daily updates, subscribe to our newsletter by clicking here.

 

One of the joys of being an old scientist is to encourage extraordinary young ones, Levitt, an American-British-Israeli biophysicist and a professor at Stanford University since 1987, said in a recent interview with Calcalist. He might have met Barbash because his wife knew his family, but that is not enough to make him go into business with someone, Levitt said. “I got on board because his vision excited me, even though I thought it would be very hard to realize.”

Levitt, 72, and Barbash, 37, met for the first time last year. It was during that meeting that Levitt, having heard Barbash’s idea, told him to start working. “We’ll see what comes out of it,” he said. Three months ago, Levitt came to New York to meet with Barbash again. In his own words, he was stunned by Barbash’s progress.

“I advise a lot of companies,” Levitt said. But Barbash’s startup, Quantified Biology, achieved “incredibly amazing” results within a very short time frame. “Not only does he have brilliant software, he has had serious customers from the start,” he said.

That wow moment was inspired by Barbash’s creation, a lightweight software that can be installed on a personal computer, capable of dramatically shortening the time it takes scientists to analyze cells and tissue images. “Research teams use very large databases, and they pose a specific question that the database can answer,” Barbash told Calcalist. “To get that answer, they use generic processing programs. But there are two problems: these programs are massive and hard to operate, and they are very inflexible, meaning they cannot answer every question. Researchers often adjust their research questions to conform with the programs.”

Quantified Biology’s algorithms can pinpoint in each image the specific information needed, Barbash explained. Programmers can write a great algorithm, but when a computational biologist writes a program for a researcher and both people speak the same language, the result is on a different level, he added. “I have studied neuroscience, and it helps me understand how the human eye perceives images. I know what the researcher sees and can translate that into an algorithm. Every line in our code can be explained, and therefore corrected.”

There are certain cells in the brain that enable electrical communication between neurons, Barbash said, offering an example. If those cells stop working, neural signals will not be strong enough to enable cognitive processes. “Pharmaceutical companies invest a lot into understanding why those cells stop working,” he said. That’s where Quantified Biology’s software comes in. A researcher can look at an image of those cells, and score them between one and 10 depending on their function. Then Barbash and his team create a code for each score that can be used for image analysis.

While existing programs can already be applied to an image and score each cell according to the same scale, researchers have no way of knowing how the program reached its conclusions, Barbash said. “It’s a black box. To understand, you need to conduct further research. But our program gives you precise answers in one go.” In machine learning, for example, Barbash said, a model can be inputted with 1,000 images of healthy tissue and another 1,000 of unhealthy tissue. After a learning process, it can analyze new images and tell the researcher whether they are healthy or unhealthy, but there is no way to understand what exact element of each image made the model decide how to classify it.

“Our program can explain its process because it listens to the researchers,” Barbash said.

One of Barbash’s customers worked with microglia, a type of macrophage cell that act as the first line of active immune defense in the brain. They are in charge of disposing of toxic material such as plaques, which are protein fragment deposits that accumulate between neurons. “A major query in neurodegenerative disease research is why microglia do not fix damages caused by disease,” Barbahs said. “A lot of researchers are trying to understand what happens to those cells; for example, whether they behave differently in different areas of the brain,” he said. As part of the research process those cells are bombarded with hundreds of thousands of molecules. Researchers can identify the cells and the plaque that damages them, and Quantified Biology can quantify the cells’ function.

“Current software requires training by inputting huge amounts of data collected by researches over a long time,” Barbash said. “The first researcher we worked with needed six months to analyze a set of data, and our software provided her with answers within a day. It can cut her research time down from years to days. The main difference is that we provide specific answers to the questions.”

The difference between the software available today on the market and Quantified Biology’s targeted software is also what caught Levitt by surprise. Today, such programs can accurately remember inputted information, but they don’t learn from it, he said. “In many experiments, there are a lot of objects that are very hard to recognize—different types of cells, moving cells, dead cells.” Barbash, he said, realized that math can be used to simplify the identification of ambiguous objects.

“Say you are working towards a drug that is meant to kill specific cells,” Levitt said. “You test it on a slide under a microscope and see how the cells react. Shahar’s (Barbash) software scans the cells and classifies them quickly and accurately. That’s a very significant shortcut.” It is a different approach, he explained. “It is more important to work on precise problems your customers are facing than to look for an algorithm that solves everything, because there is no such thing.”

Looking at Barbash’s early life, his trajectory is not clear-cut. His father is Israeli Oscar-nominated film and television director Uri Barbash. As a teenager, Barbash attended Thelma Yellin, a well-known arts high school in Israel, played the drums, and worked on his father’s film sets. But in the eleventh grade he opened a physics book and realized there was a simple way to calculate when and with what force a ball thrown into the air will come down. “I understood that everything can be explained, and solved, with math,” he said.

Barbash obtained his PhD in computational neurobiology at the Edmond and Lily Safra Center for Brain Sciences at the Hebrew University of Jerusalem. Four years ago, he arrived in New York for a postdoc at The Rockefeller University, and took part in a project set up by Teva Pharmaceutical Industries Ltd., intended to shorten the timeframe for developing drugs for neurodegenerative diseases. He was dreaming of finding a cure for Alzheimer’s disease, a condition his grandfather suffered from.

“We found more connections between disease components, we found connections between Alzheimer’s and other diseases,” Barbash said. “People understand the word ‘cure’ as something that either works or doesn’t. But during my three years at Teva’s project I learned that the process is sisyphic. It is strife with failure, and requires an amount of work that is almost inhuman and cannot be performed by academic institutions, only by pharmaceutical companies. In an academic institute you find a connection and study it for three years, but the thought of how to turn it into a drug is on the backburner.”

That is why he made the decision to leave academia and become an entrepreneur, he said. “I want to do something, anything, that will change the situation for neurological pharmaceuticals, because they are light years behind every other domain.”

Drug companies are bending over backwards to find something that could mitigate the symptoms of degenerative diseases, because there is so much failed research, Barbash said. More sophisticated tools are required to analyze brain-related research. It takes place at the microscopic level. If you cannot adequately identify the change that happens in the cells, you will not be able to identify a molecule that could be the basis for a successful drug, he said.

Today’s microscopes are good enough to create terabits of excellent information, Barbash said, meaning the bottleneck is no longer data collection but data analysis. And that exact bottleneck is what Quantified Biology wants to tackle.

Quantified Biology currently employs two people in addition to Barbash, the CEO. It has no big investors. But it already has several significant clients, thanks to Runway, a startup postdoc program operated by Cornell Tech, a New York-based joint venture between Cornell University and Israeli research university the Technion Israel Institute of Technology. Only four postdocs are chosen each year for the program, with participating startups receiving an investment of $277,000 in return for a 6% stake, and also legal and business mentorship.

“I went through 10 interviews,” Barbash said. “It was fatiguing but so worth it. We’re a for-profit company, being supported by a nonprofit.”

Barbash is a classic example of the need to find a new way to help scientists become entrepreneurs, said Fernando Gomez-Baquero, a nanomaterials engineer and entrepreneur and the director of Runway, who is also on Quantified Biology’s board of directors alongside Levitt. “He is another brilliant scientist that came up with an idea that answers a real need, but the university had no way to help him.”

Quantified Biology’s first customer was from one of the biggest players in the domain—pharmaceutical company Roche. “She had a lot of images, she had a question, and she couldn’t reach an answer.” Barbash created a specialized software for her within a week. After receiving the software, the researcher told her colleagues about Barbash, solidifying the company’s reputation and creating momentum for it.

“That was when we understood that we had something good going on, and we stopped to do it right, sign a contract, register a patent,” Barbash said.

Quantified Biology’s personalized software costs between $50,000 and $150,000 for the first year, with an option to extend the license for another $20,000-$50,000. He is not looking for an exit, Barbash said, though he is not ruling out the possibility that at some point the startup will be acquired by a company willing to let the team stay on as an independent entity.

“I want our software to help create more efficient drugs for neurological disorders,” Barbash said. “Not just degenerative conditions, but schizophrenia and bipolar disorder. The brain is my biggest love, and if we can offer even a glimmer of hope, that will be my holy grail.”