By Meghan Chua
Arezoo Movaghar earned her master’s degree in computer science and artificial intelligence. She built models based on the plentiful data found in medical records. So, when she came to UW–Madison as a PhD student and joined a research group, it surprised Movaghar to find out just how much data researchers in other fields collect.
In particular, she has been focused on a series of rich data sets collected at the Waisman Center.
“I was shocked that they are collecting all of this data – nobody ever told me about that!” she said.
Movaghar, a graduate student in Biomedical Engineering whose work spans multiple disciplines, uses computers to understand neurodevelopmental disorders such as autism spectrum disorder and fragile X syndrome by finding the patterns that are associated with them.
The project Movaghar joined as a new student sought to identify individuals who carry a fragile X premutation. Expansion in the fragile X Mental Retardation 1 (FMR1) gene causes fragile X premutation, a genetic condition affecting millions of people globally. People who carry the fragile X premutation have higher risk for neurodegenerative disorders, infertility, and having a child with disability.
However, Movaghar said that the premutation is underdiagnosed and most carriers are unaware of their condition. The genetic tests that can confirm whether an individual carries the premutation are costly and time-consuming, but knowing whether they carry the premutation can help someone better understand why they experience certain symptoms. It also informs people about their reproductive health and allows them to build a more informed health plan.
Thus, the research team wanted to find an easier, more cost-effective way to identify carriers. The project was fundamentally multidisciplinary from the beginning, with researchers including Movaghar’s advisor and biomedical engineering assistant professor Krishanu Saha, social work professors Marsha Mailick and Jan Greenberg, biostatistics and medical informatics professor David Page, and communication sciences and disorders assistant professor Audra Sterling. All of Movaghar’s faculty advisors are affiliates of the Waisman Center.
In its work, the group relied on a previous discovery by Sterling and Mailick that individuals with the fragile X premutation exhibit more language dysfluencies, such as pauses or repetition in their speech, and tend to use shorter sentences.
Armed with this information, Movaghar put her artificial intelligence training to the test – to great results.
“By using machine learning we were able to develop a method to identify premutation carriers – based on just five minutes of speech – with high accuracy,” she said.
Movaghar added that a genetic test is still required to confirm the premutation. However, the speech test (if replicated) allows researchers to find people who are likely to have the premutation and focus on genetic tests for those individuals. Movaghar recently presented her research to the Waisman Center Board of Visitors.
Artificial intelligence and machine learning played a critical role in the study, which was published in June 2017. With the droves of data that researchers collect throughout their work, it can be hard to make sense of all of the variables, Movaghar said.
“Some information can’t be discovered by traditional statistical methods because they usually only look at one variable, but in machine learning, we look at all of variables and their interactions together and it enables us to find some interesting hidden patterns,” she said.
Movaghar, who is from Iran, got started on this path by balancing her interests in medicine and computer science. She credits her mother, who is a nurse, and aunt, who is a computer scientist, for inspiring these interests. Combining the two and using artificial intelligence, Movaghar saw that she could have a bigger impact in her field. She is now pursuing a special graduate committee degree in Biomedical Informatics.
I think many people in computer science are just not aware of how much of a contribution they can have if they start working with people from these fields, especially biomedicine and social sciences. –Arezoo Movaghar
Her work integrating machine learning into identification of the fragile X premutation opens the door for the research team to develop a user-friendly mobile app to streamline data collection. Movaghar said participants will answer a few simple questions and record their voices for five minutes, then an algorithm will process the data to assess premutation status.
“Incorporating mobile devices into the research provides exciting opportunities,” Movaghar said. “We can scale up our research beyond geographical boundaries, track and monitor participants, and optimize the use of clinical resources.”
Movaghar’s advisor Saha said that as a researcher, Movaghar sees the big picture and thinks through from process to application to the impact the tools would have on patient families.
“Her training has allowed her to contribute to all of these other computational problems in ways that I was never trained to think about these problems in biostatistics and machine learning, but I think it’s where the field is evolving,” Saha said. “In particular a lot of the research projects here at the Wisconsin Institute for Discovery are trying to take advantage of some of the progress that’s been made in machine learning and computer science. These projects are linking up well with similar and complementary efforts all across campus.”
Movaghar’s current project builds on the phenotyping of fragile X premutation, looking at electronic health records to identify more potential symptoms of FMR1 gene expansion. The frameworks developed by the research group can be customized to study other genetic conditions and neurodevelopmental disorders, Movaghar said. The Waisman Center’s Lifespan Family Research Program, of which Movaghar is a graduate student member, recently published two studies in May. One focuses on the lifetime health problems experienced by deceased individuals with Autism Spectrum Disorder, and Movaghar used machine learning to differentiate their lifetime health problems from those experienced by deceased nondisabled members of the general population. In the other, Movaghar conducted the machine learning on clinical data to characterize health profiles of 3 different genetic subgroups of fragile X premutation carriers.
Having now experienced the impact her expertise can have on research across disciplines, Movaghar underscored the importance of computer scientists collaborating with researchers in other fields.
“I think many people in computer science are just not aware of how much of a contribution they can have if they start working with people from these fields, especially biomedicine and social sciences,” she said.
These projects were supported by the National Institute of Aging (R01 AG08768), the National Institute of Child Health and Human Development (R01 HD082110, P30 HD03110-40-S1), and the National Institute on Deafness and Other Communication Disorders (R03 DC011616). Support also comes from Marshfield Clinic Research Institute, the Wisconsin Alumni Research Foundation (WARF), the Waisman Center (U54 HD090256, P30 HD03352), and the Graduate School at UW–Madison.