Research Spotlight: Using machine learning to look at psychiatric conditions with professor Jinbo Bi

Jinbo Bi is a professor of Computer Science and associate head of the CSE department at the University of Connecticut. Photo courtesy of CSE department.

Researchers at the University of Connecticut are using artificial intelligence to evaluate psychiatric disorders.

Jinbo Bi is the Frederick H Leonhardt Professor of Computer Science and Associate Head of the Computer Science & Engineering Department at UConn. Bi’s research applies machine learning to the life sciences. In particular, Bi uses computers to study psychiatric conditions such as depression and addiction.

Bi explained that there are multiple scales at which a researcher can look at human biology. Psychiatric disorders typically have heterogeneous causal factors, and the biological pathway causing an individual to develop such a disorder is complicated. Although at the molecular scale, researchers have tried to identify genes that underlie the risk, it has been difficult. 

“There are different scales in biology: from the molecular scale, cellular scale, and tissue scale, to the system scale, which is human behavior,” Bi said. “Depression and addiction, they’re often characterized as behavioral disorders. Human behaviors, such as, binge drinking or withdrawal symptoms, are used to clinically characterize how severe the disorder is.”  

However, people who are diagnosed with the same disorder may have very different behavior.   

“For a group of people who are all diagnosed with depression, you see some of them can’t sleep, some of them sleep a lot, some of them don’t have appetite to eat, some eat a whole lot. So they have a heterogeneous manifestation of the disorder,” Bi said. 

Bi’s research takes into account this heterogeneity to examine the genetic factors that may predispose an individual for such a disorder. 

“In several of our publications, we examine the genomic information. The single nucleotide polymorphisms we call SNP, information of SNP markers is extracted from a person’s DNA. So we look at the genetic information we have, millions of these genetic markers,” Bi said. 

Bi integrates the genetic-level information with behavioral-level information to better understand the variations in the population that has a specific disorder. 

“We have the data for around 10,000 people who have been assessed for addition-related behavior,” Bi said. 

Beyond that, Bi will also look at MRI images of the brain to evaluate the structure and connections in the brain from another large database of over 500,000 people. 

Bi uses machine learning to evaluate all of these discrete data points as a whole. Bi gave the example of trying to find trends in genomic data or brain imaging which could point to a more homogeneous pattern of behavior in alcohol use disorder. 

“We can identify a group of people who are using alcohol in certain ways. For instance, if a group of people use alcohol less frequent than some others who are addicted to alcohol, but they drink a large amount at each drinking occurrence, it forms a specific alcohol use pattern,” Bi said. “For this group of people, if we can also identify that these people also carry a particular genetic variant, or a particular brain connectivity patterns from other people. “

Bi got her start by doing a PhD dissertation in machine learning. But, she wanted to find a practical application for her research, so she worked with Siemens to study diagnostic images for lung and colon cancer. Then she came to UConn. 

During her interview with UConn, Bi talked to a psychiatrist at UConn Health who was working on addiction at the time. 

“So he showed me how they look at the data they use. The maximum number of variables they could use was like ten variables,” Bi said. “When I looked at it. I was like, oh, he did not only have over 3000 clinical variables but also a large amount of genetic markers. How do we look at all of them together? How do we have a more quantitative approach?”

Bi said this approach was helpful because machine learning helps draw conclusions from large swaths of chaotic data. While many of these disorders do not fit into neat boxes, Bi’s research helps give a more comprehensive picture of the disorder. 

“In many regards, the world is very continuous, very quantitative. The world is not binary.”

Leave a Reply