Researchers have identified 2,500 new genes that may be implicated in autism, using a machine learning technique that analyzed the entire human genome looking for genes with the correct characteristics.
The findings, published in the journal Nature Neuroscience, vastly expand the number of autism candidate genes, and will help autism researchers pinpoint and characterize the genes that underlie the disease.
Until now, isolating the genes that cause autism spectrum disorder (ASD) has been like finding needles in a haystack. Between 400 and 1,000 genes are thought to cause the disorder, but the 65 that are currently known have taken decades to find.
“There’s a huge part of the molecular genetic basis that’s still unknown,” says Olga Troyanskaya (Princeton University), a senior fellow in the Genetic Networks program and senior author of the paper.
To identify these new genes, her team used information about 594 previously known ASD genes to teach a computer system what a typical ASD gene looked like.
The computer system they used combined software that had already been developed in the machine-learning community with a functional map of the brain generated from thousands of genomics datasets. It identified the top characteristics that ASD genes shared.
The graph shows how the genes implicated in autism group into functional clusters. The clusters provide clues to which cellular functions might be disrupted by ASD-associated genes. (Credit: Nature Neuroscience)
Using this blueprint of the typical ASD gene, the software system then analyzed the entire human genome. It flagged every gene that fit the ASD-gene profile — 2,500 of the most likely candidates.
To test whether their predictions were accurate, the team checked their newfound genes against a limited number of genes found in a study of 2,517 autism patients. Several genes they had shortlisted had indeed been identified in real autism patients.
The findings are a leap forward for autism research. Using this gene shortlist, researchers can now zero in much faster on the genes that cause the disease.
However, once autism researchers have identified a gene to study, they still need to find out what it does.
To provide this context, the researchers created a functional map that clusters the candidate genes into nine major functional categories, including synaptic transmission, sensory perception and circadian rhythm.
Some of these are related to known characteristics of the disease, such as difficulties with learning and social communication and repetitive patterns of behaviour.
The researchers also mapped where and when the ASD genes were active, and found that the disease may start early in fetal development, affecting brain structures including the cerebellum and the striatum.
Understanding how the role of ASD genes — across the brain and throughout development — will help researchers find better ways to diagnose and treat ASD, says Troyanskaya.
“The brain network is powerful in enabling generation of specific, testable hypotheses,” she says. “At the end of the day, these predictions will have to be experimentally and clinically tested, but it gives you a powerful framework for exploring these questions.”