In 2022, while on holiday in Goa, medical biotechnologist Timothy Jenkins stumbled upon a scientific preprint about artificial proteins that could stick like superglue to specific molecules. The paper came from David Baker’s lab at the University of Washington and was work that would later contribute to Baker’s 2024 Nobel Prize in Chemistry. For most readers, it would have been just another interesting scientific advance. But for Jenkins, who had spent years trying to develop better snakebite treatments at the Technical University of Denmark, it sparked an idea that could transform how we treat one of the world’s most neglected health crises.
Jenkins wondered if AI could design proteins that would clamp onto snake venom toxins. Snakebites are a problem in many countries including in India, where the “Big Four” – the spectacled cobra, common krait, Russell’s viper, and saw-scaled viper – cause thousands of deaths each year. Current treatments rely on a dangerous and costly process that hasn’t changed much in over a century.
To produce traditional antivenom, scientists must first “milk” venomous snakes, a process Jenkins describes as “like handling a live hand grenade.” The venom is then injected into horses or sheep to produce antibodies. It’s expensive, time-consuming, and requires careful refrigeration, which is a challenge in the remote areas where snakebites are most common.
Jenkins promptly fired off an email to Baker, not really expecting a response. That casual message, from a beach holiday, would lead to a breakthrough collaboration. Using the AI tool RFdiffusion, they designed custom proteins that could neutralize some of the deadliest components of snake venom- the three-finger toxins that can paralyze muscles and stop victims’ hearts.
Snake venom is a complex cocktail of dozens to hundreds of compounds evolved to be devastatingly effective. The three-finger toxins (3FTxs) are particularly tricky to neutralise. These proteins attack the body in multiple ways, shutting down critical nicotinic acetylcholine receptors that control muscle function while simultaneously destroying tissue. They’re stabilised by an intricate network of disulfide bridges that form their characteristic three-finger shape, making them both potent and difficult to counter. Traditional antivenoms often struggle to neutralise these toxins because they have limited immunogenicity- meaning they don’t trigger a strong antibody response in the animals used to produce antivenom.
But the AI approach offered a new way forward. In an interview with BBC’s Roland Pease, Jenkins offered a striking analogy as to the benefits: traditional drug discovery is like searching for a needle in a haystack, but using AI to design new proteins is like creating the exact needle you need from scratch.
While evolution has produced remarkably effective molecules over millions of years, this new approach allows scientists to design specific solutions to precise problems. RFdiffusion was able to design proteins that could bind to and neutralise different types of 3FTxs- including short-chain and long-chain α-neurotoxins, as well as cytotoxins.
The team specifically targeted the regions where these toxins normally bind to cell receptors, creating custom-built proteins that could block these interactions. The AI approach proved remarkably efficient. The team used this computational power to first simulate the behaviour of their designed proteins, then experimentally screened just the most promising candidates. It is a far cry from the early days of computer-based protein design when scientists could create novel proteins but struggled to give them practical real-world functions.
“Because the proteins were created entirely on the computer using AI-powered software, we dramatically cut the time spent in the discovery phase,” Jenkins noted. The resulting antitoxins were “easy to discover using only computational methods, cheap to produce, and robust in laboratory tests,” as Baker put it. The real test came in the laboratory. When mice were given a lethal dose of cobra toxins followed by the designed proteins, every single mouse survived. For Jenkins and Baker’s team, it was an electrifying moment.
In their Nature paper highlighting the discovery, first-author Susana Vázquez Torres, Baker, Jenkins and their colleagues demonstrated that their designed proteins were remarkably stable at high temperatures and could be produced using simple bacterial fermentation, which will be convenient to use in rural areas.
Of course, the results must hold up in more than mice and have to go through clinical trials before they can be used in people, but the implications go far beyond snakebites. The combination of artificial intelligence and protein design could open new frontiers in treating other neglected tropical diseases that have been overlooked because they primarily affect rural and economically disadvantaged communities. Researchers in resource-limited settings might be able to design targeted treatments without the massive infrastructure traditionally required for drug development.
Anirban Mahapatra is a scientist and author, most recently of the popular science book, When The Drugs Don’t Work: The Hidden Pandemic That Could End Medicine. The views expressed are personal.