[♪ INTRO].

Scientists know that autism has a genetic basis. Studies have found that genetic factors may explain anywhere from 56 to 95 percent of autism's features.

But so far, it's been really hard to identify exactly what those factors are. A new study in Nature Genetics suggests that may be because researchers have been looking at the wrong DNA. You see, earlier this week, a research team announced that they'd used artificial intelligence to uncover changes linked to autism in the stretches of DNA between genes— and that opens up a whole new avenue for autism research.

Autism Spectrum Disorder, or ASD, is a condition that shows up early in childhood and usually involves challenges with social skills, communication, and repetitive behaviors. But ASD isn't just one thing. It varies a lot from individual to individual.

While one person might have mild intellectual impairment but socialize well, another might have genius-level intelligence but struggle to communicate. So while scientists know that ASD has a strong genetic component, the specific DNA changes involved are hard to pin down. Researchers usually start by looking for changes to protein-coding genes— the sections of the genome which actually encode proteins of some kind.

And they've found plenty. Specific variants of hundreds of genes have been associated with having an autism diagnosis. But very few of them have been confirmed to be causative.

The few that have been sure bets — like, one gene is responsible, and it's passed down from parent to child — account for less than 10% of autism cases. Part of the trouble is that in most cases, ASD stems from what's known as a de novo mutation: a DNA change that a person doesn't share with their parents.. These are changes to DNA that happen in the egg or sperm cell before conception or in the early embryo during development.

But even these de novo changes to genes only seem to explain about 30% of spontaneous ASD cases— ones where both parents are neurotypical. And that's where the new study published in Nature Genetics comes in. They searched the rest of the genome for those de novo changes.

You see, despite their importance, protein-coding genes are thought to only make up about 1% of the human genome. The rest is what was once called junk DNA, because it didn't seem to do anything. Of course, now we know it's totally not junk— and a lot of it is actually super important because it controls the activity of genes.

Like, it provides attachment places for proteins called transcription factors which, as the name implies, speed up or slow down transcription—the first step in gene expression. And though these sequences are called “non-coding”, some do code for RNAs which play important roles in the processing steps that occur after a gene sequence is transcribed. And that means, to really understand the genetics of autism, scientists have to search through the DNA junk drawer.

The new study isn't the first to try, and the work on this front has been promising; it's just that it's difficult. It's pretty straightforward to find DNA sequence differences between people. But just because there's a difference doesn't mean it's causing anything.

No two people have identical DNA, and it's hard to tell the difference between a sequence change that does nothing and one that leads to a specific feature or condition. When you look for gene changes tied to specific conditions, you can narrow your search by using the genetic code to figure out if a change actually alters a protein's sequence. But we don't have a similar way of knowing whether a change to non-coding DNA will have an effect.

So when studies have tried to look for changes in non-coding DNA, they've ended up with too much noise in their datasets to tell if there's anything really going on. That's why, for the study published Monday, researchers tried a new way to sift through this messy genetic data: artificial intelligence. The team trained a neural network on more than 2,200 different DNA-regulating features, so it could predict whether changes to a particular section of DNA would likely make a functional difference.

They also trained it on a large set of known non-coding DNA changes and their effects. That way, it could generate a predicted impact score for each change it encounters— basically, how likely it is that a given change alters gene expression in a way that might lead to autistic features. And then, they fed it genetic data from actual people.

This data came from the Simons Simplex Collection, which is a repository of whole genomes from nearly 1800 families, each that has a child with autism, a neurotypical sibling, and two neurotypical parents. And, perhaps not surprisingly, they did find significant associations between ASD and changes to non-coding sequences— in their dataset, almost as many cases were linked to non-coding changes as coding ones. As you might expect, a lot of these changes were predicted to alter gene expression in brain tissues.

One thing that really stood out was how many of these changes seemed to affect the expression of genes after the transcription stage— when the messenger RNA is processed and readied for translation. As of yet, there hasn't been a lot of research on that part of the picture, so the researchers say those processes should be examined more closely. And they also found that many of the non-coding changes affected the same pathways and even the same genes as coding changes that are linked to autism.

So, it's possible that studying coding and non-coding changes together might be an easier way to zero in on the most influential genes— and that, in turn, might lead to a better understanding of how autism arises in the first place. The researchers say that in the future, this kind of analysis might even let scientists link non-coding changes to specific ASD characteristics. But the team says their immediate next steps involve refining and expanding the algorithm itself.

They say this is the first real evidence that non-coding changes can underlie complex human conditions. And being able to accurately predict how these changes affect gene expression could help unmask the hidden genetic components of all sorts of conditions— basically, all those things we know are “heritable” but can't trace to single genes. Which is a lot of things.

An unbelievable amount of information could be right in front of us just waiting to be uncovered - written in overlooked stretches our genomes. Thanks for watching this episode of SciShow News! If you like taking a deeper dive into the science behind the headlines, be sure to tune in every Friday for our news episodes.

And if you want to get notified whenever we post a video, hit that subscribe button and click the little notification bell [♪ OUTRO].