Marina Sarris

Date Published: April 29, 2026

Autistic people can be so different from each other, in their challenges, abilities, and strengths. Are there different types of autism and, if so, how many?

Scientists have been researching this question for years. One research team has proposed an answer by using artificial intelligence (AI) to study almost 5,400 children and teenagers in the SPARK autism study.¹

Using machine learning, a type of AI, the researchers analyzed information about the youth’s behavior, developmental histories, and mental health symptoms. They also examined genetic data for the children, who were ages 4 through 17.¹

Researchers came up with four types of autism:

• Social/behavioral subtype: These children had higher levels of disruptive and repetitive behaviors, attention-deficit/hyperactivity disorder (ADHD) and anxiety symptoms, and social communication problems. They tended to meet developmental milestones, like walking, on time. About 37 percent of the youth fit into this group.
• Mixed autism spectrum disorder with developmental delay subtype: These children, who made up 19 percent of the total, had significant delays in language, development, and learning. They were less likely to have symptoms of ADHD, anxiety, and disruptive behavior than youth in some of the other groups.
• Moderate challenges subtype: Children in this group had fewer difficulties with learning, social skills, repetitive and disruptive behaviors, and psychiatric symptoms than the other groups. They made up 34 percent of the total.
• Broadly affected subtype: Ten percent of the children faced the most challenges in all areas studied, including behavior, learning, psychiatric symptoms, and developmental delays.

How Was the Research Done?

Scientists from several universities and the Simons Foundation, which funds the SPARK study, used machine learning to analyze data on the autistic youth. For comparison, they included almost 2,000 of the children’s siblings, who do not have autism.¹

The researchers started with the answers to 239 questions from surveys that the children’s parents completed for SPARK. Rather than focusing on the total scores for each survey, the computer analyzed all the responses together to find patterns and groupings.¹

“The beauty of it was using all the individual items on the surveys, everything from, ‘does your child like to play the same piece of a song on repeat’ to questions about self-injury and conversation skills,” explains one of the researchers, LeeAnne Green Snyder, Ph.D., a clinical psychologist. “Each person in the study basically had their own profile,” she says. The computer analysis found four groups that each had certain traits in common that clustered together.

That does not mean that every child in a group had every feature of that subtype. “With four groups, you can imagine that there are people who don’t fit perfectly into one group or another. But if their probability of being associated with one group versus another was higher, then they were put into that group to continue with the analysis,” Green Snyder explains.

Alycia Halladay, Ph.D., the mother of a teenager in SPARK, says she believes her daughter could fit into two of the groups. “I think these groups are close approximations. This may not be perfect, but I think most parents can see some of this [traits of the groups] in their kids.”

The research team checked its findings by analyzing youth from another autism study. They found the same four groupings.¹

What About Genetics?

Researchers also analyzed the youth’s DNA to look for changes from small to large. Everyone has variations in their DNA code. Many variants have no effect, but others have small to large effects.

They wanted to know: Do members of an autism subtype have similar genetics?

Some children had rare, significant variations in genes that are known to cause autism and developmental disorders. Many of these variants were new to the child, meaning that they were not inherited, while some were passed down by a parent. These are the kinds of variants that SPARK will tell participants about, if they wish to know.

The researchers also looked for small, common variations in DNA. By itself, a common variant does not have a noticeable effect. But if a person has certain groups of these variants, they can add up to create a “meaningful effect,” said research team member Natalie Sauerwald, Ph.D., of the Flatiron Institute for Computational Biology at the Simons Foundation.²

Scientists can add up these common variants to create a polygenic (meaning many genes) score. This score can give an estimate of a person’s chances of having a higher IQ or ADHD, for example, based on the presence of variants linked to these traits. But these scores do not show whether the person actually has a trait or condition.

For the study, researchers calculated polygenic scores for autism, ADHD, depression, IQ, and epilepsy among autistic children and their siblings.¹

Do People With the Same Type of Autism Have Similar Variants?

According to the study, autistic children within a subtype tended to have similar polygenic scores, on average. They also had similar rates of new and inherited variations in genes that have moderate to significant effects on brain development.

For example, the social/behavioral group overall had higher polygenic scores for ADHD and depression. They also had more symptoms of those conditions. These children were more likely to have variants in genes that are active after birth.

Children in the broadly affected group were more likely to have new, rare variants in genes that have significant effects on development. They also had the most challenges in all areas that were studied.

The youth in the mixed autism with developmental delay group had a mix of inherited and new variants, particularly in genes active before birth. Like the broadly affected group, they had significant delays in language, development, and learning.

The moderate challengesgroup, which had fewer overall difficulties, tended to have variants in genes that have a smaller effect on development.

Is an Autism Subtype the Same as a Diagnosis?

Sauerwald and Green Snyder say that the four subtypes are not a diagnosis. “Your doctor cannot give you a formal diagnosis according to these groups,” Green Snyder explains.

Also, the study did not look at whether autistic adults fit into the four groupings or whether children changed groups as they got older, Green Snyder says. Those questions need to be answered in future research.

Researchers and doctors continue to look for better ways to describe the types of autism. “This was not the first, and not the last, attempt at doing this. There have been thousands of studies and papers talking about subtypes of autism,” Green Snyder says.

In the beginning, studies divided autism into just two categories: people who have intellectual disability or limited language, and people who do not. Previous versions of the psychiatric diagnosis manual also divided autism into several diagnoses, often referred to as classic autism, atypical autism, and, for children without speech delays, Asperger syndrome. But those categories were not very useful. People with similar traits ended up with different autism-related diagnoses.

So autism experts combined all previous diagnostic categories into one, autism spectrum disorder, in 2013.

The Importance of Defining Types of Autism

Sauerwald and Green Snyder say it’s possible that future studies, with larger numbers of participants, may find more than four subtypes.

Green Snyder says that she has worked with autistic children who might be a subgroup within the broadly affected group. “These kids have a much higher nonverbal IQ, who can run circles around me doing puzzles and visual spatial tasks, and they cannot talk,” she says.

Alycia Halladay, an autism expert who was not involved in this study, says that finding subtypes can help target the right support to autistic people.

“This is a way for researchers to better understand what particular supports and what particular trajectories people are likely to have, rather than just lumping everyone together with autism, which doesn’t seem to be very useful in helping the right people get the right support at the right time,” says Halladay, chief science officer of the Autism Science Foundation.

SPARK participant Denise Lombardi, parent of an autistic son, says that finding subtypes is important “to gain clarity on the biology of autism and ensure no one is left behind.”

Interested in joining SPARK? Here’s what you should know.

Photo credit: iStock

References

1. Litman A. et al. Nat. Genet. 57, 1611-1619 (2025) Article
2. Sauerwald N. SPARK webinar. Accessed Apr. 14, 2026.