A researcher’s innovative mathematical model for analyzing brain imaging has unveiled connections between genes, brain structure, and autism.
A collaborative research team from multiple universities, co-directed by engineering professor Gustavo K. Rohde from the University of Virginia, has created a system that can effectively detect genetic markers associated with autism in brain images, achieving an accuracy rate of 89 to 95%.
The results indicate that in the future, doctors could diagnose, categorize, and treat autism and similar neurological disorders using this technique, eliminating the need to depend on or wait for behavioral indicators. This means the potential for personalized medicine could lead to earlier intervention efforts.
“While autism is traditionally identified through behavioral assessments, it has a significant genetic component. Adopting a genetics-first approach could revolutionize how we comprehend and treat autism,” the researchers mentioned in a paper published on June 12 in the journal Science Advances.
Rohde, a professor in biomedical as well as electrical and computer engineering, teamed up with scholars from the University of California San Francisco and the Johns Hopkins University School of Medicine, including Shinjini Kundu, who is Rohde’s former Ph.D. student and the paper’s lead author.
During her time in Rohde’s lab, Kundu, who is currently a physician at Johns Hopkins Hospital, contributed to developing a generative computer modeling technique known as transport-based morphometry (TBM), which serves as the foundation for the team’s method.
The team’s system leverages a unique mathematical modeling technique to identify patterns in brain structure that correlate with variations in specific regions of a person’s genetic makeup—aspects referred to as “copy number variations” where segments of genetic material can be deleted or duplicated, which are associated with autism.
TBM enables the researchers to differentiate between typical biological variations in brain structure and those connected to genetic deletions or duplications.
“Some copy number variations have known associations with autism; however, the relationship between these variations and brain morphology—how different types of brain tissues, such as gray and white matter, are organized—remains largely unexplored,” stated Rohde. “Understanding how CNV connects to brain tissue structure is a crucial first step to grasping the biological foundations of autism.”
Understanding TBM’s Mechanism
Transport-based morphometry differs from other machine learning-based image analysis techniques because it relies on mathematical models grounded in mass transport—the flow of molecules like proteins, nutrients, and gases in and out of cells and tissues. The term “morphometry” means measuring and analyzing the biological shapes produced by these processes.
According to Rohde, most machine learning techniques do not account for the biophysical processes that produced the data and instead focus on identifying patterns to detect anomalies.
In contrast, Rohde’s method extracts mass transport data from medical images through mathematical equations, generating new images for visualization and more in-depth analysis.
Next, by employing a separate set of mathematical techniques, the system isolates information related to autism-associated CNV variations from regular genetic variations that do not cause disease or neurological issues, which the researchers refer to as “confounding sources of variability.”
These confounding factors have previously hindered researchers from comprehending the relationship between genes, brain structure, and behavior, thereby confining healthcare providers to behavior-based assessments and treatments.
As per Forbes magazine, 90% of medical information is derived from imaging, which remains largely untapped. Rohde holds the belief that TBM could serve as a crucial key to unlocking this data.
“As a result, many significant discoveries from such extensive data could be forthcoming if we apply more suitable mathematical models to extract this information,” he explained.
The team utilized data from participants involved in the Simons Variation in Individuals Project, which consists of subjects with genetic variations linked to autism.
Control subjects were recruited from various clinical environments and matched for characteristics such as age, sex, handedness, and non-verbal IQ, while ensuring exclusion of those with related neurological disorders or family histories.
“We hope that our findings on identifying specific changes in brain structure linked to copy number variations could illuminate brain regions and mechanisms that can be targeted for therapies,” Rohde stated.
Publication
The study focusing on the gene-brain-behavior relationship in autism through generative machine learning was published on June 12, 2024, in Science Advances.
Additional co-authors include Haris Sair from Johns Hopkins School of Medicine and Elliott H. Sherr and Pratik Mukherjee from the Department of Radiology at the University of California San Francisco.
Funding for this research was provided by the National Science Foundation, the National Institutes of Health, the Radiological Society of North America, and the Simons Variation in Individuals Foundation.
