Bioengineers have used artificial intelligence to resurrect molecules from the past by identifying fragments of Neanderthal proteins that have the ability to fight bacteria. This groundbreaking study, published July 28 in the journal Cell Host & Microbe, may inspire new drugs to treat human infections, writes Nature.

The researchers applied computational methods to data on the proteins of both modern humans (Homo sapiens) and our long-extinct relatives – Neanderthals (Homo neanderthalensis) and Denisovans. This allowed them to identify molecules capable of killing disease-causing bacteria.

“We’re motivated by the notion of bringing back molecules from the past to address problems that we have today,” says Cesar de la Fuente, a study co-author and bioengineer at the University of Pennsylvania in Philadelphia.

Antibiotic development has slowed over the past few decades, and most of the antibiotics prescribed today came on the market more than 30 years ago. As the number of antibiotic-resistant bacteria increases, a new wave of treatments will soon be needed.

Many organisms produce short protein subunits called peptides that have antimicrobial properties. Some of these antimicrobial peptides, mostly isolated from bacteria, are already used in clinical practice. Proteins of extinct species can become an untapped resource for the development of antibiotics.

The researchers trained an artificial intelligence algorithm to recognize places on human proteins where they are known to be cleaved into peptides. They applied this algorithm to publicly available protein sequences of H. sapiens, H. neanderthalensis, and Denisovans. Using the properties of previously described antimicrobial peptides, they predicted which ones might kill bacteria.

At the same time, finding and testing drug candidates with the help of AI takes only a few weeks, compared to three to six years using old methods to discover one new antibiotic.

The researchers tested dozens of peptides to see if they could kill bacteria in a lab dish. They then selected six peptides – four from H. sapiens, one from H. neanderthalensis and one from Denisovans – and injected them into mice infected with the bacterium Acinetobacter baumannii, a common cause of infection in humans.

All six peptides stopped the growth of A. baumannii in the thigh muscle, but none of them killed the bacteria. The five molecules killed bacteria growing in skin abscesses, but at “extremely high” doses, according to Nathanael Gray, a chemical biologist at Stanford University in California.

Modifying the most successful molecules can create more effective versions, and changing the algorithm can improve the identification of antimicrobial peptides, reducing the number of false positives. “Even though the algorithm that we used didn’t yield amazing molecules, I think the concept and the framework represents an entirely new avenue for thinking about drug discovery,” says de la Fuente.

Euan Ashley, an expert in genomics and precision medicine at Stanford University in California, is excited about this new approach in the little-studied field of antibiotic development. De la Fuente and his colleagues have shown that diving into the archaic human genome is an interesting and potentially useful approach.