As deadly ‘superbugs’ outsmart antibiotics, a Penn scientist uses artificial intelligence to design new ones
The need for more antibiotics is urgent. Worldwide, 700,000 people die of a drug-resistant infection each year, and the number is projected to hit 10 million by 2050.
Ever since the accidental discovery of penicillin in 1928, the main strategy to find other lifesaving antibiotics has been to make that type of accident happen on purpose: Identify chemical compounds that kill bacteria in nature, and repurpose them for use in humans.
Yet in every case, bacteria have begun to outsmart these lifesaving medicines, sometimes developing resistance less than a year after the drugs are introduced. And no major new class of antibiotics has been developed in decades.
Cesar de la Fuente thinks it is time to look for new drugs with a 21st-century tactic: artificial intelligence.
“We think that nature, perhaps, has run out of inspiration,” he said.
In his lab at the University of Pennsylvania, de la Fuente and his team use computer algorithms to design and refine new drugs, sifting through chemical building blocks for combinations that are predicted to penetrate bacterial defenses.
He described one such success in a study last year while at Massachusetts Institute of Technology: a twisting, branching molecule that disrupted the membranes of bacteria called Pseudomonas in infected mice, yet left the animals’ own cells intact.
More research would be needed before trying it in humans, but in the meantime his team of biologists, chemists, and computer whizzes is at work developing more compounds — including some that could be administered probiotic-style, as part of a dietary supplement.
The need for more antibiotics is urgent. Worldwide, 700,000 people die of a drug-resistant infection each year, and the number is projected to hit 10 million by 2050, according to a review cosponsored by the United Kingdom’s Department of Health and the Wellcome Trust.
In the United States, the annual death toll is at least 23,000, according to a 2013 analysis by the U.S. Centers for Disease Control and Prevention, which is issuing an update sometime this month. A growing number of infections are caused by what are popularly called “superbugs” — bacteria that are resistant to multiple types of antibiotics.
The circumstances have long been worrisome for those with vulnerable immune systems: newborns in intensive care units, recipients of organ transplants, and patients with cancer or HIV. Lately, physicians fear the problem could emerge in otherwise healthy people who undergo routine surgery, such as a hip replacement.
“It’s enough to cause a rational person to be very, very concerned,” said Baylor University biologist Christopher Kearney, who specializes in microbial research.
Mounting a stronger attack
It is no surprise that nature has been the primary source for current antibiotics. For millions of years, plants, fungi, and other organisms evolved all sorts of ingenious chemical defenses against bacteria, so human chemists have been happy to copy and tweak them. Penicillin, for example, is produced by a type of fungus, killing bacteria by weakening their cell walls. Its discoverer, Alexander Fleming, is said to have remarked: "I did not invent penicillin. Nature did that.”
Bacteria even have evolved compounds to attack each other. But for every type of attack, whether it comes from a natural foe or doctor’s syringe, bacteria have proven equally adept at mounting a defense. They can open channels in their membranes to flush out antibiotics, for example, or produce chemicals to degrade them. And once they develop these means of resistance, bacteria can share the battle plan — the DNA — with their peers.
At Penn’s Perelman School of Medicine, de la Fuente’s search for new antibiotics starts with the building blocks of life, amino acids, yet he is combining them in ways nature never envisioned.
In a sense, it is a giant math problem. There are untold trillions of ways that the 20 core amino acids can be arranged into short molecular chains called peptides — more possible combinations than the number of stars in the universe — yet a fraction of 1% of those possibilities exist in nature. And just two have been turned into actual antibiotics. De la Fuente operates on the hope that somewhere among all those combinations, a new weapon is just waiting to be found.
But he needs electronic brainpower to find it. For the 2018 study, he and colleagues started with a type of naturally occurring antibiotic: a peptide found in the seed of the guava plant.
Naturally, this peptide has weak antimicrobial properties, but he and his fellow researchers were intrigued by two characteristics. It contained certain amino acids with a positive charge, meaning they would be drawn toward the negatively charged membrane of bacteria. And it included other amino acids whose chemistry enabled them to interact with, and disrupt, these fatty membranes.
If the scientists could randomly generate hundreds of cousins of this natural peptide on the computer — chopping it into fragments, recombining them, and swapping in different amino acids in a few locations — maybe the interplay between those two characteristics could be optimized. The peptide could mount an even stronger microbial attack.
The researchers identified the 15 most promising compounds from among the hundreds that were generated by the algorithm, made them in the lab, and tested them against bacteria. Indeed, several had enhanced antimicrobial activity, and one in particular stood out.
Dubbed guavanin 2 because of its guava-seed origins, this molecule is still a peptide. But it is dramatically different from its natural origins, with unusually high levels of an amino acid called arginine, among other distinctive features, de la Fuente said. It is not entirely clear why it works better than its natural ancestor.
Little incentive for drug companies
Among the lab’s next projects is making the process more automated, while incorporating some of the principles of artificial intelligence. The system would generate molecules and run computer simulations, then guide a robotic arm to make and test the winners against live bacteria. Data on which combinations proved most effective would then be fed back into the model for the next round.
“You essentially have a positive feedback loop that is a self-learning system and discovery platform,” de la Fuente said.
Others warn that science is just part of the solution, calling for more incentives to encourage drug development.
Big drug companies steer clear of developing new antibiotics largely because the drugs are not moneymakers, said Eili Klein, a senior fellow at the Center for Disease Dynamics, Economics & Policy, a think tank with offices in Washington and New Delhi. By nature, the drugs are designed to be used for the short term, some of them only as a last resort.
“They’re inherently self-limiting if they’re effective,” said Klein, an assistant professor at Johns Hopkins University.
What’s more, physicians are constantly reminded to use them sparingly, as each time an antibiotic is used, the bacteria that survive the treatment have the potential to multiply and spread. Judicious stewardship is a good thing, but it does not help sales.
Policy questions aside, none of it would matter without scientists to lay the foundation. And, in de la Fuente’s case, with electronic wizardry to guide the way.