By Phoebe Ingraham Renda
Illustration by Phoebe Ingraham Renda
Just as machine learning can guide our entertainment choices, University of Pittsburgh researcher Afshin Beheshti, professor of surgery in the School of Medicine, asks it to do a more serious task: predict small molecule drugs that can target microRNAs.
Small, But Mighty
MicroRNAs are small noncoding RNA molecules that orchestrate vast biological effects. They regulate gene expression by degrading messenger RNA (mRNA) molecules and getting in the way of ribosomes, the molecular machines trying to read and translate mRNAs into proteins. Some microRNAs can even bind to and silence the expression of hundreds of thousands of genes. As a result, they are key players in many biological pathways, functions and disease states.
“There are good ones you need for your body, and then there are bad ones,” says Beheshti, who directs Pitt’s new Center for Space Biomedicine and is the associate director of the McGowan Institute for Regenerative Biomedicine.
For example, cancer can develop when “good” microRNAs that silence oncogenes (genes that promote cancer) are suppressed. Conversely, cancer can also develop if “bad” microRNAs that silence tumor suppressor genes (genes that help prevent cancer) are too abundant.
“There’s a signature of microRNAs associated with every disease or pathway that causes that disease to progress,” says Beheshti. As a result, these signatures are good biomarkers for tracking disease risk and progression.
It Liked This, It Might Like That
That’s where the power of machine learning comes in. Just as streaming services use algorithms to suggest your next binge-worthy series based on your viewing history, Beheshti and colleagues developed a tool that predicts which microRNAs a drug might interact with by analyzing the chemical similarities to the drug’s known biological targets—in other words, its biological “watched list.” To help accelerate the research process, which can take five to 10 years for new therapies, Beheshti and colleagues focused on repurposing FDA-approved small molecule drugs to target microRNAs.
Small molecules can selectively bind to microRNAs and influence their activity; however, identifying an approved drug that might target a microRNA of interest from a database of thousands is no small task—at least for humans. With the expertise of Diego Galeano, a professor of engineering at the National University of Asunción in Paraguay, and the power of machine learning, Beheshti and Galeano created sChemNET to identify drug candidates.
“Using this machine learning tool we're able to show that if you have a microRNA, or a list of microRNAs, you can put it into the tool and it'd say, ‘here's a list of drugs that could potentially target the microRNA,’” says Beheshti.
To validate sChemNET’s suggestions, Beheshti and colleagues at Boston University, Morehouse School of Medicine and Purdue tested the small molecule drugs using in vivo and in vitro experiments. Using human cell lines, fish and mammalian models, they saw that sChemNET’s drug suggestions reversed the biological damage caused by dysregulation of miR-451, miR-181 and others, which are involved in conditions like anemia, cancer and myocarditis.
Getting a Step Ahead
Beheshti notes that as more microRNA-disease associations are discovered, incorporating microRNA-biomarker screening into annual clinical check-ups could allow for early disease detection. Developing small molecule drug therapies for these early-stage disease-associated signatures could be a first step toward preventing disease progression.
“For some cancers, like lymphoma, you're not going to know you have it until you feel something wrong with your body, and at that point, the cancer is harder to treat,” says Beheshti. “If you could stop it early on, potentially using this tool, that cancer could never develop—that’s the dream.”