ASU researchers uncover the rules that guide how microRNAs control genes

When a cell produces a microRNA, it creates two unique strands from a single molecule. For years, scientists have known that only one of those strands usually becomes active, but they have not understood how the cell makes that choice.

A new study from Arizona State University shows that the decision is not random. Instead, it follows consistent, conserved rules that can be identified and predicted using artificial intelligence.

The research, published in Nucleic Acids Research, answers a long-standing question in molecular biology and reveals a previously hidden layer of gene regulation. By combining large-scale experiments with machine learning, the team demonstrates that microRNA strand selection is governed by features encoded in the RNA itself and shaped by evolution.

“We knew microRNAs were powerful regulators of genes,” said Marco Mangone, professor in the Biodesign Institute and the School of Life Sciences at ASU. “But we didn’t understand how cells decided which strand to use, and that choice completely changes which genes are regulated.”

MicroRNAs are short RNA molecules that regulate gene expression by binding to messenger RNAs and preventing them from being translated into proteins. Each microRNA forms a hairpin structure that is cut into two strands. Although both strands contain regulatory information, typically only one is loaded into the cell’s gene-silencing machinery.     

Which strand is selected matters because each targets a different set of genes. Choosing one strand over the other can redirect regulation across hundreds, and sometimes thousands, of genes.

“If you use one strand, you regulate one network of genes,” Mangone said. “If you use the other, you regulate a completely different network.”

Understanding that choice is especially important because microRNAs play key roles in development, stress responses and disease. Changes in microRNA processing or strand usage have been linked to cancer, neurodegenerative disorders and other complex conditions.

To uncover the rules behind strand selection, the researchers turned to the model organism Caenorhabditis elegans, a microscopic nematode whose genetic systems are highly conserved with humans.

Led by molecular and cellular biology PhD student Dalton Meadows, the team developed an high-throughput method called HiTmiSS that allowed them to precisely measure which microRNA strands were used across developmental stages and tissues.

Rather than studying a handful of examples, the team generated tens of thousands of strand-specific measurements, creating one of the most comprehensive datasets of microRNA strand usage to date.

“We needed enough data to move beyond individual cases,” Meadows said. “The goal was to see whether there were consistent rules underlying strand selection.”

Once the experimental data were in place, the challenge became identifying patterns that no human could reasonably extract by eye.

The team worked with Assistant Professor Heewook Lee from the School of Computing and Augmented Intelligence and the Biodesign Institute’s Center for Biocomputing, Security and Society to develop a machine learning approach to analyze the dataset. Using this approach, they identified 77 sequence and structural features that together predict which strand will be selected. These features include nucleotide composition, pairing mismatches and subtle differences in RNA structure.

The resulting predictive model accurately determined strand usage in worms. When the researchers applied the same model to human microRNAs, they were surprised to find that it still performed well, despite hundreds of millions of years of evolutionary divergence.

“That told us we weren’t just seeing a worm-specific phenomenon,” Meadows said. “We were seeing something fundamental.”

The findings suggest that while microRNAs have evolved to become more precise in mammals, the underlying logic guiding strand selection has been preserved across species.

The study establishes microRNA strand selection as a regulated and predictable process rather than a random one. That insight has broad implications for how scientists study gene regulation and disease.

A single change in strand usage can shift the regulatory targets of a microRNA, potentially disrupting entire gene networks. Being able to predict strand selection from sequence alone could help researchers better interpret genetic data, understand disease mechanisms and develop RNA-based diagnostics.

“If people take one thing away,” Mangone said, “it’s that cells follow rules we’re only just beginning to understand, and now we finally have the tools to see them.”

Importantly, the rules uncovered by this framework are deeply conserved across species and directly relevant to human biology, as abnormal microRNA strand selection has been implicated in cancer, neurodevelopmental disorders and other diseases.

The team has made their data and predictive tools openly available so other researchers can apply the framework across species and biological contexts. Future work will explore how strand selection rules have changed across evolutionary development and how they may be altered in disease.

Together, the findings highlight the power of combining experimental biology with artificial intelligence to reveal hidden principles of life and underscore ASU’s growing role at the intersection of biology and computation.