Israeli team uses tool that finds fake online profiles to detect abnormal protein activity
Ben-Gurion University scientists say their innovative algorithm, WGAND, can identify rogue protein behavior the same way investigators uncover suspicious social network patterns
Reporter at The Times of Israel
In an intriguing study, a Ben-Gurion University of the Negev cybersecurity researcher who analyzes fraud on social networks joined forces with a team of BGU biologists to develop a machine-learning system to recognize abnormal activity in protein networks inside the human body.
Their innovative method, weighted graph anomalous node detection (WGAND), uses an algorithm that uncovers suspicious behavior in social networks such as LinkedIn or Instagram to discover anomalous behavior in networks of proteins inside cells.
The researchers said WGAND enabled them to identify proteins associated with brain disorders and heart conditions, as well as those involved in critical biological processes, like neuron signaling in the brain and muscle contraction in the heart.
“It’s exciting to see how bringing together expertise from cybersecurity can lead to breakthroughs in understanding human biology,” said Dr. Michael Fire, assistant professor in the Software and Information Systems Engineering Department at the university, who worked with lead researcher Dr. Esti Yeger-Lotem, associate professor in the Department of Clinical Biochemistry & Pharmacology, Dr. Juman Jubran and Dr. Dima Kagan.
The study was recently published in the peer-reviewed journal GigaScience.
From opposite ends of the campus
Speaking to The Times of Israel by telephone, Fire said he and Yeger-Lotem “are on opposite ends of the BGU campus” and hadn’t met until the university announced it was offering grants for joint research projects.
“There’s a strong effort to encourage interdisciplinary collaboration, including grants for researchers who work together across faculties,” said Fire. “I work with people from other fields because AI has become an integral part of many different domains.”
He was curious about how he could use his research in AI in computer science to help biologists explore interactions of proteins “like a social network,” he said.
Fire explained that in his work, he identifies atypical patterns among users in social networks to uncover fraudulent transactions or fake profiles.
The people on online social networks who try to “steal your money or send you viruses are probably using a fake profile to do it,” he said. “Most of them are finding the victim randomly by connecting to many communities and many groups.”
In contrast, Fire said, regular users are most often connected via a small number of groups.
By creating an algorithm to predict the links between two users, he can find fraud and other irregularities. From this concept of connections in social networks, he said, “We move to networks in biology.”
Tracking down suspicious behavior of proteins
Proteins have been dubbed the workhorses of biochemistry. Known as nature’s versatile tool, these essential molecules in the body interact with one another in complex networks, called protein-protein interaction (PPI) networks.
Scientists can understand how proteins function and how they contribute to health and disease with an analysis of their networks.
This is where Yeger-Lotem’s work comes in.
In her lab, she develops and applies novel computational approaches in network biology, studying how proteins, genes, and other molecules communicate, and treating them as if they belong to a large social network inside the human body.
The same algorithms that uncover irregularities in social networks can be applied to atypical behavior in the networks of proteins.
The analysis of the interaction patterns among proteins, Yeger-Lotem said, can uncover which proteins play special roles, both positively and negatively, in tissues such as the brain and heart.
“Proteins don’t act alone,” Yeger-Lotem told The Times of Israel by telephone. “Basically, like any molecule, they act by interacting with other molecules. So we look at protein interactions and ask why they seem different in one tissue than in another.”
The WGAND algorithm can help researchers identify which genes and processes are important in different tissues and why certain diseases happen, she said.
‘A generic algorithm’
While there are other ways to study protein interactions, the researchers said that WGAND outperformed existing methods in terms of accuracy and precision.
“What is really cool about our method is that it is a generic algorithm,” Yeger-Lotem explained. “We can use it for predicting interesting protein behaviors, and in the same way, we can predict fake profiles or changes in a medical or transportation network.”
WGAND is open source, allowing researchers worldwide to utilize and build upon it.
“Everything is open, and you can use it, and it’s meant to be really easy to use,” she said. She encourages scientists to “send us an email and say that they used it and found something interesting.”
The two researchers are already working on their next project, Yeger-Lotem said.
“It’s not always easy, but it’s a fun collaboration,” she said.
The Times of Israel Community.