Meet the Autism Data Science Initiative grantees

Mathematician Amy Cochran, assistant professor of population health sciences and mathematics at the University of Wisconsin-Madison, has proposed to employ casual inference methods to find factors—for example, an increase in parental age—that explain the rise in autism diagnoses. Cochran and her co-investigator Maureen Durkin, professor of population health sciences and pediatrics at the University of Wisconsin-Madison, also aim to examine whether any factors they identify as contributing to autism likelihood also play a role in life outcomes.

The team plans to pull data from the Study to Explore Early Development (SEED) cohort and electronic health records from UW Health. Casual inference methods are well suited for this type of analysis of many variables, because they enable researchers to flexibly explore many different relationships, Cochran says. “It’s this good balance of like, OK, we can learn things that we might have not learned using traditional statistics approaches, but we’re going to try to protect against overfitting so we don’t get these spurious associations.”

The apparent rise of autism prevalence in the United States means that there will soon be more autistic adults—an increase of up to 300 percent by 2030, based on the latest estimate of 1 in 31 children having a diagnosis, says Laura Klinger, associate professor of psychiatry at the University of North Carolina at Chapel Hill and executive director of the university’s TEACCH Autism Program. “I know we’re not ready to think about what those services and support needs might be for autistic adults and for their families.”

Klinger proposed a project to investigate how various factors contribute to quality of life for autistic adults, including community participation, mental and physical health and the use of assistance programs. The data come from surveys completed by 1,400 autistic adults—or their caregivers if they could not complete the survey themselves—at six National Institutes of Health (NIH) Autism Centers of Excellence, as well as a cohort of 400 autistic adults from the Simons Foundation Powering Autism Research (SPARK) study. (SPARK is funded by the Simons Foundation, The Transmitter’s parent organization.)

Investigators at the six centers had already been discussing a potential collaboration before the ADSI was announced, Klinger says. She is leading the project alongside Lindsay Shea at Drexel University and Gregory Wallace at George Washington University; the other participating centers are Children’s National Hospital, the University of Pittsburgh and the University of Wisconsin-Madison. The ADSI grant enables Wallace, one of the SPARK investigators, to follow up with the SPARK cohort and collect data at a third time point, Klinger says.

Kristen Lyall is an epidemiologist who has studied how maternal pregnancy complications, chemical exposures and diet relate to autism. The project she proposed unites those interests, says Lyall, associate professor in the Modifiable Risk Factors Program at Drexel University’s Autism Institute. “I call it the ‘diet and’ study.” She says she plans to use existing data from the NIH’s Environmental influences on Child Health Outcomes (ECHO) program, a collection of longitudinal studies that explore how genetics and the environment affect child development.

One element of the project focuses on whether pesticides and other chemicals in a woman’s prenatal diet track with the likelihood of having an autistic child. In another arm of the study, Lyall says she intends to examine whether prenatal diet affects how strongly a pregnant parent’s metabolic health shapes their child’s likelihood of being diagnosed with autism.

Lyall plans to collaborate with Wei Perng, associate professor of epidemiology at the Colorado School of Public Health, to identify metabolomic pathways that might mediate any associations the team uncovers. That effort will help the team assess “if we are seeing a signal in these association analyses, does it make sense, and is there a biologic link?” Lyall says. Finally, she aims to investigate whether autistic children experience worse nutrition than non-autistic children, and whether differences in diet contribute to co-occurring conditions or trait severity.

Discussions of how prenatal factors contribute to autism can come across as blaming parents, Lyall says, but her team’s goal is to “really push for policy changes to alter exposures, rather than placing the onus and the burden on the individual, because a lot of these exposures as an individual we can’t control.”

Since the early 2000s, the Children’s Hospital of Philadelphia (CHOP) has consistently screened children for autism during well-child visits, says Judith Miller, associate professor of psychiatry at CHOP. As a result, CHOP has a cohort of more than 4,000 children with autism and 100,000 without. Miller plans to combine that dataset with the health records of the children’s parents and geocode data that links zip codes to specific environmental exposures, such as poverty levels, air and water quality, and access to green spaces and early interventions. Her team also plans to examine exposure to policy changes, such as changes to the autism diagnosis criteria in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, and a 2024 Pennsylvania law requiring insurers to cover autism interventions.

“Anything that the child has experienced can count, can be considered an exposure. So it doesn’t have to be just environmental toxins,” Miller says.

From there, the team aims to perform risk prediction modeling to identify variables that may contribute to getting an autism diagnosis and to determine their statistical weights. A community panel of autistic people, researchers and clinicians will help interpret the clinical meaning and relevance of any identified variables, Miller says.

The team plans to upload a de-identified version of the cohort’s data to the NIH’s database of Genotypes and Phenotypes (dbGaP), which will be the first time that credentialed investigators who aren’t affiliated with CHOP will have an opportunity to work with the data, Miller says.

Epidemiological studies of autism to date have been limited to pulling out only correlations, because they could not incorporate genetic data, says Jonathan Sebat, professor of psychiatry at the University of California, San Diego. Some studies have tried to control for genetic influence by including siblings, but that does not offer insight into the mechanistic interaction between genes and the environment, he says.

The project he proposes corrals several biobanks and large cohort-based studies to create a database of 2.7 million people. The data sources include the Adolescent Brain Cognitive Development (ABCD) Study, the HEALthy Brain and Child Development (HBCD) Study and the Simons Foundation Powering Autism Research (SPARK) study, among others.

By examining environmental exposure and genetic information from the same people in such a large cohort, the team intends to “tease apart” casual pathways, Sebat says. “It’s a new era. That wasn’t possible in the past.”

Gene-by-environment studies aim to untangle how these two factors combine to determine a person’s responses to potentially toxic chemical exposures. But it can be difficult to measure such exposures in epidemiological studies, and even more difficult to separate the effects of a single exposure from their combined effects, says Jason Stein, associate professor of genetics at the University of North Carolina at Chapel Hill. So, for greater control, Stein says he wants to do “gene by environment in a dish.”

Stein and Joseph Piven, professor of psychiatry and pediatrics at the University of North Carolina at Chapel Hill, generated 115 induced pluripotent stem cell lines from participants of the Infant Brain Imaging Study. Stein and his colleagues plan to produce organoids from the cell lines, expose them to environmental substances and perform single-cell RNA sequencing to see if the exposures affect gene expression or other molecular processes. He also intends to explore whether the organoids’ different genetic backgrounds of the organoids contribute to any measured effects. The team plans to collaborate with Heather Volk, professor of public health at Johns Hopkins University, to examine whether any gene-environment interactions that affect the development of organoids are also relevant in people.

His team’s proposed list of chemical exposures features hydrocortisone, to mimic in-utero exposure to the steroid as a result of maternal stress; methyl mercury, to model doses of the heavy metal from eating certain fish; perfluorooctanoic acid, a model of prenatal PFAS exposure; the heavy metal cadmium; the plasticizer phthalate; and the pesticides chlorpyrifos and deltamethrin. But that list will likely evolve based on the interests of other ADSI groups, Stein says.

One of the many complications involved in studying a chemical exposure’s effect is that exposures do not act in isolation: “We’re exposed to thousands of things throughout our lifetime, and those can interact with other factors and our biological systems,” says Douglas Walker, associate professor of environmental health at Emory University.

To capture some of that complexity, Walker and his team propose to characterize thousands of small molecules in blood samples from parents and children enrolled in the Early Autism Risk Longitudinal Investigation (EARLI) and the SEED cohorts—specifically from pregnant women who went on to have a child with autism, and from autistic children, non-autistic siblings and parents of a child with autism. From there, they plan to identify environmental exposures associated with autism and whether those relationships vary with and across families. “We’re essentially taking an omics-based approach to study our environment, which has been desperately needed,” Walker says.

The group intends to assemble two datasets: a database of chemicals that are associated with neurodevelopment or neurotoxicity, created through computational and web-scraping methods, and the “Autism Exposure Atlas,” containing exposures and their biological response.

This project is one of two efforts to replicate, validate and reproduce findings from the 11 other groups. Biostatistician Judy Zhong and her colleagues developed Results Reproduction (R-squared), a statistical model replication pipeline. The team has used the pipeline for more than two decades to replicate more than 130 datasets, says Zhong, professor of population health sciences at Weill Cornell Medicine.

The project provides three services: If one of the other ADSI groups finds an association between a factor and autism, Zhong’s team will attempt to replicate it using the portion of data the group set aside when developing their model. Zhong’s team will also try to validate the model by testing it on data from a different source. And they plan to work with each team during the model development phase to make sure they document their methods in a way that makes it easy for other researchers to reproduce. “We feel that is so important for building trust in the whole scientific community,” Zhong says.

Research on positive life outcomes for autistic youth often does not factor in what the youth themselves say makes for a happy life, says Katharine Zuckerman, professor of pediatrics at Oregon Health & Science University School of Medicine. “So the first part of the project is that we’re going to do that.”

Zuckerman’s team plans to assemble a panel of autistic youth and adults, their parents and their health-care and education providers, and ask them what elements make up a good life. The panel will also help the researchers identify relevant variables in databases—such as Medicaid claims, the National Survey of Children’s Health and early-intervention data from several U.S. states—that can be used to model those outcomes. Similar projects in the past have had to choose a particular outcome to study, such as graduation rates, but the ADSI enables Zuckerman and her team to “go big” and take a “global perspective,” she says. At the end of the project, the panel can also help turn any identified factors into policy and program recommendations.

Zuckerman says she hopes her team “can provide [HHS with] some information that says, ‘Hey, if you really want to improve the lives of autistic people, these are the things your own data say will do that,’”—and that the HHS will invest in those things.