For millions of Americans living with chronic eye diseases, preserving sight can mean a demanding daily eye drop routine or recurring, uncomfortable injections directly into the eye.
Now, researchers at the University of Maryland and the Johns Hopkins Wilmer Eye Institute are turning to artificial intelligence, specifically machine learning, to develop a potential alternative: long-lasting eye drops that could reduce the need for frequent dosing and invasive procedures.
Supported by a new $2.35 million grant from the National Eye Institute, part of the National Institutes of Health (NIH), the team is using machine learning to design multifunctional peptides—short chains of amino acids engineered to perform several tasks in delivering medication to the eye. Led by Michael Cummings, an MPower Professor of biology with an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS), the initiative aims to improve how drugs are delivered to the eye.
According to the Centers for Disease Control and Prevention, glaucoma, one of the primary targets for this delivery approach, is a leading cause of irreversible blindness in the United States. More than three million Americans are living with the condition. That figure is projected to reach 6.3 million by 2050 as the population ages. The condition also costs the U.S. economy an estimated $2.86 billion annually in direct medical expenses and lost productivity.
Because glaucoma has no cure, long-term management focuses on lowering pressure inside the eye. But following a daily eye drop regimen over decades can be difficult. For other chronic eye diseases, treatment may require frequent injections, while implantable drug delivery systems carry their own risks of adverse reactions.
To address these challenges, the interdisciplinary team is developing methods that release medication gradually over time. Laura Ensign-Hodges, vice chair for research and director of the Center for Nanomedicine at the Wilmer Eye Institute, will direct the preclinical laboratory research to evaluate drug properties, partnering with Cummings to combine data science with medicine.
“Two aspects of the project I find personally appealing,” said Cummings, who directs UMD’s Center for Bioinformatics and Computational Biology. “The first is the potential to positively impact one aspect of human health. The second is the interesting elements of computer science we can apply.”
The researchers’ approach centers on ocular melanin, a natural pigment in eye tissue that can act as a reservoir to slowly release medication. Because many therapeutic drugs do not naturally bind to melanin, researchers in Cummings’ lab use computational modeling to engineer peptides that can serve as adaptors. These custom peptides must perform three tasks: bind to melanin, penetrate eye tissue, and be non-toxic.
To find the most effective designs, Cummings’ team uses machine learning to simulate a process similar to natural evolution. Starting with a pool of candidate peptides from earlier lab work, the computational model evaluates and scores each variant on key performance measures—what Cummings called its “fitness.”
“We subsequently select the top performers and repeat this scheme several times, each time improving the peptide characteristics,” he said.
Once the computational modeling identifies the strongest candidates, Cummings passes them to the Johns Hopkins team of bioengineers, pharmacologists and molecular biologists for laboratory testing.
“My collaborators at Johns Hopkins then evaluate the top candidates in the laboratory for effectiveness and safety,” Cummings said. “It is an excellent example of wet lab and computer lab collaboration, with machine learning in the middle, rather than at the end as is typically the case.”
The initiative builds on findings published in Nature Communications in May 2023. In that initial study, Cummings and his collaborators demonstrated the approach by engineering a peptide adaptor consisting of 12 amino acids. When attached to the glaucoma medication brimonidine, the peptide bound to retina pigment and extended the drug’s retention in the eye from eight days to as long as 18 days. The team subsequently received a U.S. patent for their work.
“The new project expands this previous research to longer peptides that we hope will be even more effective,” Cummings said.
By designing these longer amino acid chains to bind more strongly, minimize toxicity and penetrate cells more effectively, the team aims to establish a versatile drug delivery platform. If successful, it could support longer-lasting eye drops for glaucoma and potentially replace injections or surgical implants for other conditions, including ocular neovascular diseases, which involve abnormal blood vessel growth in the eye.
—Story by Melissa Brachfeld, UMIACS communications group