AI detects regions around 'microexons' involved in genetic variation in autism
New research from scientists is helping piece together the missing genetics of autism spectrum disorder (ASD).
In a study published in Nature, researchers identified hundreds of new microexons — short gene segments that appear to play a key role in ASD and other neurological conditions — and developed a machine learning approach to predict the impact of ASD-associated genetic variants on the regulation of these segments.
“Autism is characterized by multifactorial genetics,” says the study’s co-senior.
“Hundreds of genomic variants have been associated with autism, and these deliver a genetic diagnosis in 15 per cent of autistic individuals,” says another author.
That means that for 85 percent of people with an ASD diagnosis, the genetic explanation for their condition remains unknown. The authors believe that many contributing genetic factors may have remained undetected because they occur in regions of the genome that do not have a known function.
The genomic regions surrounding microexons are one such example. Through a process known as alternative splicing, these very short snippets of DNA — which range from three to 27 DNA “letters” long — can be included or excluded from the messenger RNA instructions used to make proteins. The resulting proteins vary slightly in composition yet contribute to a greater diversity of critical protein functions.
In the newly published study, the authors used a computational tool called MicroExonator to analyze over 100 terabytes — enough storage for 25 million smartphone photos — of publicly available RNA sequencing data and identified more than 1,500 new microexons, of which nearly 300 appeared to be specifically spliced in the brain.
The researchers also observed that many of the newly discovered microexons are concentrated in genes associated with critical brain-specific processes, further supporting the idea that these tiny DNA fragments contribute to neurological health.
The researchers then asked whether DNA variants found in autistic individuals might preferentially impact the expanded repertoire of microexons.
To answer that question, the researchers build and applied a machine learning model that identifies the key regulatory features governing whether a microexon is spliced in or out of messengers RNAs. This so-called regulatory code encompasses specific sequences of DNA letters flanking microexons.
“What surprised me was that the sequences the model identified as important closely matched those conserved throughout vertebrate evolution” says the author, noting that DNA sequences that are highly conserved, or the same, between species generally have critical functions.
The researchers applied their model to whole genome sequence data from thousands of autistic people and neurotypical sibling pairs, and analyzed the predicted impact of ASD-associated genetic variants on the microexon regulatory code.
“Using our model, we were able to show that rare genetic variants generally have the largest predicted effects on the splicing of neuronal microexons,” says the author. The model further pinpointed a subset of rare variants predicted to alter microexon splicing that are associated with autism.
The research team also found that the variants predicted to affect microexons are predominantly detected among the 85 per cent of autistic individuals previously lacking a genetic link to autism.
These new findings build on over a decade of research from the lab to uncover the role of microexons in ASD, including earlier work showing that microexons are frequently misregulated in the brains of autistic people and that mice lacking microexon splicing display multiple ASD-like behaviours.
“In terms of future work, it will be important to understand how disrupted microexons affect interactions between proteins that are important for nervous system development and function,” says the author. One of the research programs in the lab is also exploring new therapeutic approaches to target microexon dysregulation.
The researchers believe that the machine learning model they developed can be used to characterize newly discovered genetic variants and their effects in other neurological conditions, like schizophrenia and epilepsy, where microexons are believed to play a role.
