A team of engineers has collaborated with autism researchers to develop a “machine-learning” program that scours a large but little-explored region of a person’s genome for changes associated with autism and other disorders. Already, the technique has identified several new autism-linked genes that promise insights into how the disorder develops.
The study, supported in part by Autism Speaks, appears online today in the journal Science. Its authors include Stephen Scherer, director of the Autism Speaks genomics program MSSNG. Dr. Scherer also directs the University of Toronto McLaughlin Centre and The Centre for Applied Genomics at Toronto’s SickKids Hospital.
“Massive genomic databases such as MSSNG are only as good as the tools and innovations in analysis that researchers bring to them,” says Autism Speaks Chief Science Officer Rob Ring. “Machine-learning is just the type of analysis necessary to realize the full value of the genomic data we’re generating. We love to see this happening here.”
In essence, machine-learning enables a computer to sort through a huge amount of information to look for meaningful patterns. In this case, the program is combing through more than 60,000 variations in the human genome to predict which are most likely to be associated with autism symptoms and medical conditions. (See image above.) More specifically, it zeroes in on the little-explored region of the genome that directs the splicing of messenger RNA. This type of RNA translates DNA into proteins. Proteins, in turn, are involved in all aspects of development and body function.
“This work is groundbreaking because it represents a first serious attempt to decode portions of that 98 percent of the human genome outside the genes that are typically analyzed in genetic disease studies,” Dr. Scherer says. “This is particularly exciting since it is thought these segments may contain much of the missing information that we have been looking for in autism research.”
Dr. Scherer has pioneered the use of genomics to yield medically useful information for individuals affected by autism and other developmental disorders. (Read more about Dr. Scherer’s work here.)
Collaborating with Dr. Scherer, the computational team compared mutations in the genomes of children with autism to those present in a group of other children. Using traditional genetic analysis, the team found no differences between the two groups. But when they used their new computations system to identify mutations that change RNA splicing, surprising patterns appeared.
In all, they identified 39 previously unrecognized gene changes associated with autism. In addition, their method uncovered gene changes related to hereditary cancers and spinal muscular atrophy.
Read more about MSSNG here.
“What we know about autism is not enough. MSSNG is the search for the missing answers.”
In the video below, University of Toronto computational scientist Brendan Frey discusses how his “machine learning” program is enabling researchers to explore the human genome in ways not previously possible – with a primary emphasis on better understanding autism.
Subscribe to Autism Speaks Science Digest for autism research news, perspective and expert advice delivered biweekly to your inbox.