AI Unlocks Antibiotic Potential in Deadly Venoms

With the help of AI, researchers uncovered antimicrobial compounds in animal venom, potentially revolutionizing the treatment of drug-resistant infections.

Snakes, spiders, and scorpions are venomous creatures that strike fear in the hearts of many. A single sting or a bite from various species of these animals is enough to cause serious damage, or even death in humans. But what if these deadly venoms held the secret to combating one of modern medicine’s most pressing threats: antimicrobial resistance?

Every year, antimicrobial resistance claims millions of lives, rendering once-treatable infections fatal.¹ The rampant and unregulated use of antibiotics has fueled the emergence of multi-drug-resistant bacteria. In the face of this growing crisis, researchers are turning to unexpected sources for solutions.

Animal venoms are one such untapped treasure trove of antimicrobials.² “Venoms are evolutionary masterpieces,” said César de la Fuentea computational biologist at the University of Pennsylvania. “They’ve spent hundreds of millions of years learning how to breach diverse biological defenses.” Over the past few years, scientists have mined the rich diversity of venom peptides for their therapeutic potential. For example, researchers originally sourced ziconotide, a painkiller, from cone snail venom, and derived captopril, a medication for high blood pressure, from snake venom.^3,4 In fact, semaglutide, the popular weight-loss drug, was designed to mimic the venom of the Gila monster.⁵ However, the potential of venoms as a source of antibiotics remains largely unexplored.

Now, in a recent study, de la Fuente and his team used artificial intelligence to comb through the vast library of venom proteins and discovered brand-new antimicrobial compounds with potent antibacterial activity. ⁶ Their findings, published in Nature Communicationshighlight the potential of venom-derived therapeutics to reduce the burden of antimicrobial resistance in the coming years.

“Venom compounds are fast acting, very potent, and very specific. All the ingredients you look for when you’re trying to make a drug,” said Mandë Holforda chemical biologist at Hunter College who researches venom peptides in marine snails and was not involved in the study. “This story demonstrates the power of venom.”

One of major hurdles to mining venoms for antimicrobials is the vast number of bioactive peptides within venom proteomes.⁷ “Each venomous animal has in its venom arsenal upwards of 200 or more unique peptides, proteins, and small molecules. That’s an astonishingly enormous number, in terms of compounds to explore,” Holford said.

To systemically identify new antibiotics within the complex web of venom proteins, de la Fuente and his team relied on a deep learning model that could predict function from protein sequence. They sourced over 16,000 venom proteins from four databases and fed their amino acid codes into the model. “It sort of operates like a barcode scanner,” de la Fuente explained. “It scans a proteome and tells you which region is likely to be a good antibiotic.” Learning from tons of data on established antibiotics, the model predicted the presence of over 40 million venom-encrypted peptides (VEPs) in the venom proteins with antibiotic potential.

To further narrow down the list of potential antibiotics, the researchers compared the structural and functional properties of VEPs to known antimicrobial peptides (AMPs). They weeded out peptides with high similarity to AMPs and ended up with 386 candidates. Many of the VEPs were more positively charged than AMPs, which could facilitate their interaction with the negatively charged bacterial membranes and amplify their bactericidal activity.

“With AI systems that we developed recently, the cool thing is the scale and the rapid time frames of what we can achieve,” de la Fuente said. “We can now discover new compounds in a few hours, whereas before it would take years.”

Next, the team tested the antimicrobial activity of a subset of the identified VEPs by incubating them with cultures of 11 clinically relevant pathogens. Out of the 58 VEPs they analyzed, 53 showed potent activity against at least one pathogenic strain of bacteria.

To understand the mechanism by which VEPs acted on bacterial cells, the team tested their effect on two things: permeabilization and depolarization of the bacterial membrane. The peptides primarily acted by membrane depolarization.

Finally, de la Fuente and his group analyzed how well VEPs performed in an animal model. They infected skin abscesses in mice with the pathogenic bacteria Acinetobacter baumannii and treated the wounds with VEPs. Three compounds showed promising antimicrobial activity, with a single topical dose reducing bacterial counts two days post infection. None of them affected the animal’s weight, indicating minimal toxicity.

Now, de la Fuente and his team are trying to chemically modify the identified VEPs to improve their stability and potentially translate them into the clinic. “Hopefully, other researchers (will) also join the efforts to explore venoms as a source of potential therapeutics, not only for infectious diseases, but also for many, many other things,” he said.