Machine Learning Tool Pinpoints Crucial Gene Regulation Changes

Changes in our genes affect health risks from heart disease to cancer, but how those genes are turned on and off is just as crucial to health. What's more, almost all genes occur as two copies-one from each parent-that aren't necessarily turned on and off in sync. When that differential gene regulation goes awry, it can cause serious health problems. 

But until now, it's been difficult to detect copy-specific differences in gene regulation and understand which of them are important.

Now, researchers have developed a tool that pairs machine learning with advanced DNA sequencing to disentangle differential gene regulation between maternally and paternally inherited chromosomes and find regions that are crucial for gene function.

Their new tool, called FIRE, could help researchers pinpoint the DNA changes behind hereditary diseases-a key step toward genetic medicines-and provides an unprecedented look at previously mysterious regions of the genome.

"Only about 1 or 2% of our genome codes for proteins," says Mitchell R. Vollger, PhD, assistant professor of human genetics at University of Utah Health and co-lead author on a study describing the results. "This gives us a much sharper lens on the other 98%, which includes the switches that turn our genes on and off."

The team also discovered that lasting changes in gene regulation aren't always triggered by genetics. Two identical copies of the same stretch of DNA can be regulated differently, and those differences are passed down as our cells divide. The function of most of these newly discovered changes is unknown, but they suggest a new dimension of biological variability that isn't determined by genetics.

The results are published in Cell.

This Is Your Genome on FIRE

The researchers used FIRE to reveal how changes in DNA sequence affect the regulation of health-critical parts of the genome. 

Previous studies had found a genetic difference that was associated with risk of celiac disease, but no one knew if the difference actually affected how genes work or was just an innocent bystander-an important distinction for developing genetic medicines. 

The celiac-associated genetic difference sits in a gene "switch" next to an immune gene that has also been linked to celiac disease, the researchers found. The DNA change disrupts a spot where gene-controlling proteins normally attach, and changing just that one DNA letter in lab experiments was enough to change how open the switch was. This suggests that, out of thousands of nearby genetic differences, this is likely the single one that changes gene regulation.

Using FIRE, the team also identified tens of thousands of previously overlooked DNA regions that are likely to be important for gene regulation. These regions are more likely than other parts of the genome to contain DNA changes associated with human disease, which suggests that they have important, undiscovered functions for health.

The Limits of Genetics

FIRE also enabled researchers to search on an unprecedented scale for parts of DNA that are regulated differently between maternally and paternally inherited chromosomes. What they found was unexpected. In most cases, differences in gene regulation between the two copies of a DNA region could be linked to differences in the DNA itself. But in a large minority of cases, the two differentially regulated copies were genetically identical at that spot.

I think the most surprising result of this work was finding the limits of how much our genetics can determine our gene regulation. About 40% of the time that we see a regulatory difference between the copies we inherited from mom and dad, there's no genetic difference between them at that spot."

Mitchell R. Vollger, PhD, assistant professor of human genetics, University of Utah Health

How FIRE Works

FIRE starts with an existing technique that "highlights" physically accessible regions of DNA, then uses advanced sequencing to read long stretches of individual DNA molecules. Because each stretch is so long, the researchers can usually tell whether it came from the mother's or father's copy based on slight differences in the sequence. Highlighted regions can indicate active genes or parts of the DNA that are important for gene regulation-but they can also mark less-important regions that just happen to be accessible to the "highlighter" at that moment.

The new tool adds a machine learning-based analysis that can check every one of these highlighted regions and assess which ones are likely to be actually important for gene regulation. Researchers can do a similar assessment by hand for individual sites, but the tool scales it up to check billions of highlighted regions across the genome.

Machine learning here is really copying something that we're able to see with our own eyes in the data. But it's doing it at a scale that we could never manage by eye.

Mitchell R. Vollger, PhD

Toward Cures for Genetic Diseases

Vollger hopes that, by linking changes in the genetic code to functional differences in gene regulation, FIRE will help find which genetic differences cause human disease-a crucial early step toward developing therapies that edit DNA to treat or even cure hereditary diseases.

"This might show us how we could edit the DNA switches that control a person's genes to help lower their risk for a common disease, or even cure them of a rare disease," Vollger says. "It's a small step toward turning the genetic variants we find into something doctors can act on."