summary: Using a combination of machine learning and neuroimaging data, researchers report being able to classify people on the autism spectrum into four different subtype groups based on their brain activity and behavior.
sauce: Weill Cornell University
A study by researchers at Weill Cornell Medicine found that people with autism spectrum disorders can be classified into four different subtypes based on their brain activity and behavior.
Research published March 9 Nature Neuroscienceused machine learning to analyze newly available neuroimaging data from 299 individuals with autism and 907 neurotypical individuals.
They found patterns of brain connectivity associated with behavioral traits in people with autism, such as language ability, social influence, and repetitive or stereotypic behavior.
They confirmed that the four subgroups of autism could also be reproduced in separate datasets, and showed that differences in regional gene expression and protein-protein interactions explained differences in brain and behavior.
“As with many neuropsychiatric diagnoses, individuals with autism spectrum disorders experience varying types of difficulties with social interactions, communication, and repetitive behaviors.
“Scientists believe there are probably many different types of autism spectrum disorders that require different treatments, but there is no consensus on how to define them.” Feil Family Brain in Weill Cornell Medicine and Mind Research Institute.
“Our research sheds light on new approaches to discover subtypes of autism, which may one day lead to new approaches for diagnosis and treatment.”
Previous research published by Dr. Liston and colleagues natural medicine In 2017, we used similar machine learning techniques to identify four biologically distinct subtypes of depression, and subsequent studies showed that these subgroups responded differently to various depression treatments. It was shown to show.
“By sorting people with depression into appropriate groups, we can assign them the most appropriate treatments,” says lead author Amanda Buch, Ph.D., postdoctoral fellow in neuroscience in psychiatry at Weill Cornell Medicine. said.
Based on that success, the team set out to determine whether similar subgroups exist among individuals with autism and whether different genetic pathways underlie them. , explained that autism is a highly heritable condition, linked to hundreds of genes, which present with diverse symptoms and limited treatment options.
To investigate this, Dr. Buch pioneered new analyzes to integrate neuroimaging data with gene expression data and proteomics, introduced them into the lab, and explored how risk variants might affect subgroups of autism. allowed us to test and develop hypotheses about how
“One of the obstacles to developing therapeutics for autism is the wide range of diagnostic criteria, which are therefore applicable to large, phenotypically diverse populations with different underlying biological mechanisms. That’s it,” said Dr. Buch.
“It is important to understand and target this biodiversity in order to individualize treatment for people with autism. It is difficult to identify an effective treatment.”
Until recently, there was not a large enough collection of functional magnetic resonance imaging data from autistic patients to conduct large-scale machine learning studies, Dr. Buch noted. But the large datasets created and shared by Dr. Martino, Dr. Adrianadi, research director of Child’s Mind His Institute’s Center for Autism, and other colleagues across the country are not enough for research. provided a large dataset.
“New methods in machine learning that can process thousands of genes, differences in brain activity, and multiple behavioral variations made this study possible,” said co-lead author, Weill Cornell Medicine, Department of Psychiatry. said Dr. Logan Grosenick, assistant professor of neuroscience. Pioneering machine learning techniques used for biological subtyping in autism and depression research.
These advances allowed the team to identify four clinically distinct groups of people with autism. Two of the groups had above average verbal intelligence. One group also had significant deficits in social communication, but less repetitive behavior, while the other group had more repetitive behavior and less social impairment.
Connections between parts of the brain that process visual information and help the brain discern the most important incoming information were much more active in the more socially impaired subgroup. It was weaker in the group with positive behavior.
Dr. Book, Ph.D., from Weill Cornell Graduate School, said: He completed his PhD in Medical Science in Dr. Liston’s lab and now he works in Dr. Grosenick’s lab.
Two other groups had severe social impairments and repetitive behaviors, but language abilities on opposite ends of the spectrum. We found completely different brain connectivity patterns in these two subgroups.
To better understand what is driving the differences, the team analyzed gene expression that accounts for the atypical brain connections present in each subgroup, many of which were previously associated with autism. They also analyzed network interactions between proteins associated with atypical brain connections, looking for proteins that might serve as hubs.
Oxytocin, a protein previously associated with positive social interactions, was a hub protein for a subgroup of individuals with high social impairment but relatively limited repetitive behaviors.
Studies have looked at the use of intranasal oxytocin as a treatment for people with autism, and the results have been mixed, Dr. Buch said. She said it would be interesting to test whether oxytocin therapy is more effective in this subgroup.
“You can get effective treatment for subgroups of people with autism, but the lack of attention to that subgroup undermines that effect in large trials,” Dr. Grosenick said.
The team checked the results on a second human dataset and found the same four subgroups. As a final validation of the team’s results, Dr. Buch noted that other studies had independently linked autism-associated genes with the same behavioral traits associated with subgroups she developed in the biomedical literature. We conducted an unbiased text mining analysis.
The team will next study these subgroups and potential subgroup-targeted therapies in mice. Collaborations with several other research teams with large human datasets are also underway. The team is also working to further refine their machine learning techniques.
“We are trying to make machine learning more cluster-aware,” said Dr. Grosenick.
During that time, Dr. Buch said he received encouraging feedback about their work from people with autism. His diagnosis was confusing because the disease was so different from others, but said her data helped explain his experience.
“Having been diagnosed with a subtype of autism may have helped him,” Dr. Buch said.
About this machine learning and autism research news
author: press office
sauce: Weill Cornell University
contact: Press Office – Weill Cornell University
image: image is public domain
Original research: closed access.
“Molecular and network-level mechanisms that explain individual variability in autism spectrum disorders,” by Conor Liston et al. Nature Neuroscience
Molecular and network-level mechanisms that explain individual differences in autism spectrum disorders
The mechanisms underlying phenotypic heterogeneity in autism spectrum disorders (ASD) are poorly understood.
Using large neuroimaging datasets, we identified three potential dimensions of functional brain network connectivity. This predicted individual differences in ASD behavior and was stable on cross-validation.
Clustering along these three dimensions revealed four reproducible ASD subgroups with distinct functional connectivity changes in ASD-associated networks and clinical presentation profiles reproducible in independent samples. rice field.
By integrating neuroimaging data with canonical gene expression data from two independent transcriptome atlases, we found that within each subgroup, ASD-associated functional connectivity correlated with regional differences in expression of distinct ASD-associated gene sets. found to be explained by
These gene sets were differentially associated with different molecular signaling pathways, including immune and synaptic function, G protein-coupled receptor signaling, protein synthesis and other processes.
Taken together, our findings delineate the atypical connectivity patterns underlying different forms of ASD that involve different molecular signaling mechanisms.