AI-Designed Proteins Show Promise in Neutralizing Deadly Snake Venom Toxins

Lethal snake venom may be countered by new AI-designed proteins

An artificial intelligence tool designs proteins that match toxins scientists want to target Artificial intelligence could take the bite out of snake venom. Using AI, scientists have designed proteins that say not so fassst to toxins wielded by cobras and other venomous snakes. It's a proof-of-concept approach that could one day offer a new treatment for snakebites. In lab experiments, the custom proteins saved the lives of mice given an otherwise lethal dose of toxins, researchers report January 15 in Nature. These proteins "are really doing their job," says Michael Hust, an antibody researcher at the Technical University of Braunschweig in Germany who was not involved with the new research. "The mice are surviving. This is what we all want." The work represents the latest real-life application of work that earned three scientists the 2024 Nobel Prize in chemistry. In 2022, medical biotechnologist Timothy Jenkins spotted a preprint from the lab of David Baker, a biochemist at the University of Washington School of Medicine in Seattle and one of the Nobel awardees. The preprint described AI-designed proteins that stick like superglue to specific molecules. That sparked an idea. Could the AI think up a design that clamps onto -- and neutralizes -- snake venom toxins? Jenkins, of the Technical University of Denmark in Lyngby, had spent years trying to develop new therapies for snakebites. Worldwide, snakebites kill some 100,000 people each year. Venomous snakes can deliver a blizzard of toxins via their bite. Some of the most dangerous include molecules called three-finger toxins, which can paralyze muscles, stilling people's hearts and their ability to breath. Antivenoms exist, but the technology is outdated, Jenkins says. "There's not a lot of money in it, so not a lot of innovation has been attracted." Current antivenom producers milk snakes to extract their venom, which is "like handling a live hand grenade," he says. A small dose of that venom gets injected into a horse or other large animal, from which producers later harvest antibodies. When given to a snakebite victim, those antibodies bind to venom toxins and shut them down. But manufacturing antivenom is costly and time-consuming, so scientists have been searching for other methods. One option that's seen recent success is scanning a vast collection of lab-made antibodies to identify those that target particular toxins. With AI, scientists can quickly and cost-effectively build toxin-targeting proteins from scratch. Jenkins and Baker paired up to create custom proteins using a generative AI model called RFdiffusion. It's a free protein-design tool that shares some similarity with the AIs that generate images. Instead of conjuring up a picture of the pope in a puffer jacket, RFdiffusion can concoct protein designs that match a molecule scientists want to target. Baker's team had previously trained the model on all known protein structures and their amino acid sequences, the string of molecular building blocks that fold up into a protein's 3-D shape. Then, the researchers computationally disassembled those shapes. That taught the model how to put together a complete protein from its components, like learning how to build a car engine by taking it apart. Baker and Jenkins asked the AI to design proteins that would glom on to venom toxins. Then they manufactured the proteins in the lab. Like a magnetic cap covering the tip of a key so it no longer fits in a lock, the synthesized proteins prevented the toxin from docking onto cells. The team injected 20 mice with the custom proteins 15 minutes after a lethal dose of cobra toxins or concurrently with the toxins. Every mouse survived. "We were very, very excited about this," Jenkins says. It was a stark demonstration of the proteins' powers. Next, the team wants to develop its proteins into an actual product it could test in people. Scientists will need to ensure the custom proteins are safe, and not binding unexpectedly in human tissues, Hust says. Jenkins agrees. The new study is just a first step to defanging venoms' harms. "It was very much just proving that this extremely new technology works," he says.

AI-designed proteins tackle century-old problem -- making snake antivenoms

Proteins designed using artificial intelligence (AI) can block the lethal effects of toxins delivered in the venom of cobras, adders and other deadly snakes. The AI-designed proteins could form the basis of a new generation of therapies for snakebites -- which kill an estimated 100,000 people each year and are still treated much as they were a century ago. The study, published in Nature on 15 January, is also a demonstration of how machine learning has supercharged the field of computational protein design. Challenges that used to take months or years, or even prove impossible -- such as designing a new protein to successfully block another -- can now be accomplished in seconds. "It's scary," says Joseph Jardine, an immunologist at Scripps Research in La Jolla, California. "It's gone from 'we couldn't even do this' to proof-of-concept work solving actual problems." In many parts of the world, snakebites are a prominent killer and cause of permanent disability. The World Health Organization in Geneva, Switzerland, has named snakebite a top-priority neglected tropical disease, alongside others such as dengue and rabies. Yet treatments have changed little in more than a century -- most are based on antibodies in blood serum taken from horses and sheep immunized with snake venom. These antivenoms vary in safety and efficacy and must be administered in a health clinic by trained staff, limiting their usefulness, notes José María Gutiérrez, a toxinologist at the University of Costa Rica Clodomiro Picado Institute in San José. Developing snakebite treatments wasn't on the radar of David Baker, a computational biophysicist at the University of Washington in Seattle, when his lab unveiled a game-changing protein-design program called RFdiffusion in late 2022. Inspired by image-generating AI tools such as DALL-E and Midjourney, the program proved adept at designing small proteins that bind strongly to target proteins, including those related to cancer and autoimmune diseases. Susana Vázquez Torres, a biochemist in Baker's lab, was interested in tackling neglected diseases and wondered whether RFdiffusion could help to improve snakebite treatments. Snake venoms are composed of various protein toxins that cause paralysis and tissue damage. Vázquez Torres, Baker and their colleagues used RFdiffusion to design 'mini-binders' that recognize key regions of three kinds of toxin made by elapid snakes -- the family that includes cobras, mambas and adders. The researchers identified mini-binders that attached remarkably strongly to snake-venom toxins after screening just a few dozen proteins for each design ("Which is crazy," Vázquez Torres notes). Further experiments in lab-grown cells showed that the mini-binders neutralized the effects of the venom components -- those that target neurotransmitters present in muscle and nerve cells, as well as those that cause tissue damage. When the researchers premixed the neurotransmitter-targeting mini-binders with their target toxins and injected mice with an otherwise lethal dose, the animals were fully protected. In experiments that better mimicked a snakebite, the researchers delivered the same two mini-binders 15 minutes after the animals were injected with lethal toxin doses, and still all the animals survived. "This is probably the coolest experimental result I've had in my career so far," Vázquez Torres says. "These antivenoms were a fantastic case for binder design. What they did is very smart," says Martin Pacesa, a structural biologist at the Swiss Federal Institute of Technology in Lausanne. It will be a long time before the AI-designed proteins can be turned into effective snakebite treatments, but they have several things going for them, say scientists. Lab-designed mini-binders tend to be extremely stable, so treatments using them probably wouldn't require refrigeration -- unlike existing antivenoms and more potent, monoclonal-antibody-based snakebite treatments in development. They could also be churned out at low cost, using bacteria in industrial fermenters. However, the toxins that Vázquez Torres and her colleagues targeted are just a few components of the venoms made by cobras and other elapids. An effective antivenom would also need to block a class of toxins called phospholipases, says Jardine. Vázquez Torres imagines that a useful antivenom would be a cocktail of mini-binders, its composition varying on the basis of the venomous snakes found in the areas where it is used. Baker's team is exploring how to move the AI-designed antivenoms to the clinic. His group has had little trouble attracting investment to apply protein design to tackling illnesses such as cancer and autoimmune diseases -- one company co-founded by Baker raised US$1 billion last year. But that kind of money isn't available to combat neglected disease. "The path forward for anything to do with infectious disease or developing-world diseases like snakebites, it's just harder," Baker says.

Researchers use AI to design proteins that block snake venom toxins

It has been a few years since AI began successfully tackling the challenge of predicting the three-dimensional structure of proteins, complex molecules that are essential for all life. Next-generation tools are now available, and the Nobel Prizes have been handed out. But people not involved in biology can be forgiven for asking whether any of it can actually make a difference. A nice example of how the tools can be put to use is being released in Nature on Wednesday. A team that includes the University of Washington's David Baker, who picked up his Nobel in Stockholm last month, used software tools to design completely new proteins that are able to inhibit some of the toxins in snake venom. While not entirely successful, the work shows how the new software tools can let researchers tackle challenges that would otherwise be difficult or impossible. Snake venom includes a complicated mix of toxins, most of them proteins, that engage in a multi-front assault on anything unfortunate enough to get bitten. Right now, the primary treatment is to use a mix of antibodies that bind to these toxins, produced by injecting sub-lethal amounts of venom proteins into animals. But antivenon treatments tend to require refrigeration, and even then, they have a short shelf life. Ensuring a steady supply also means regularly injecting new animals and purifying more antibodies from them. Having smaller, more stable proteins that perform the same function would let us produce them in bacteria and could allow the generation of an antivenon that doesn't require refrigeration -- a careful consideration given that many snake bites occur in rural areas or the wilderness. The new work isn't meant to be a complete solution to the problem. Instead, it tackles a single type of toxic venom protein: the three-finger toxins, named after the physical structure that the proteins fold into. They're a major component of the venom of such infamous snakes as mambas, taipans, and cobras. Despite their relatively compact size, different members of the three-finger toxin family manage to produce two distinct types of damage. One group causes a general toxicity to cells, enabled by disruption of the cell membrane, while a different subset has the ability to block the receptor for a neurotransmitter.

Scientists use AI to create completely new anti-venom proteins

'To find something that works in the first attempt, that's very shocking.' Each year, snake bites kill upwards of 100,000 people and permanently disable hundreds of thousands more, according to estimates from the World Health Organization. Promising new science, enabled by state-of-the-art technology, could help quell the threat. Researchers have successfully designed two proteins to neutralize some of the most lethal venom toxins, using a suite of artificial intelligence tools, per a study published January 15 in the journal Nature. These "de novo" proteins-molecules not found anywhere in nature-protected 100% of mice from certain death when mixed with the deadly snake compounds and administered in lab experiments. "I think we could revolutionize the treatment [of snake bites]," says Susana Vázquez Torres, lead study author and a biochemist who completed this research as part of her doctoral thesis in David Baker's lab at the University of Washington. Baker won the 2024 Nobel Prize in Chemistry for his work creating new proteins. This week's publication is a continuation of that line of inquiry. "This study, of course, doesn't solve the whole problem, but it demonstrates that we can develop molecules super quickly compared to traditional methods-and it works," Vázquez Torres tells Popular Science. The strategy could lead to cheaper, safer, and more effective remedies than the status quo, she adds. [ Related: Why are there so many snakes? ] "It's fantastic work," says Joseph Jardine, an assistant professor of immunology and microbiology at the Scripps Research Institute. Jardine wasn't involved in the new study, but has previously published research developing synthetic antivenoms for the same sorts of compounds. This new research is both a demonstration of how far protein design has come in recent years, enabled by rapidly improving AI, and also an exciting practical advance in medicine, he says. Despite the toll that snake bites take, the treatment for envenomings has been the same for more than a century: Antibodies collected from horses or other animals inoculated with sub-lethal amounts of venom. These antivenoms save lives, but they have some serious downsides. For one, they're expensive and difficult to make as producing them involves maintaining stables of animals. Plus, they vary in quality as relying on imperfect immune systems yields uneven results, and antivenoms tend to work better against some toxins than others-only partially neutralizing the smallest components of the complex cocktail that is venom, and performing poorly against some species' bites. They can trigger allergic reactions and other adverse side effects in recipients. And, because they're a biological product, traditional antivenoms are very sensitive to temperature and need to be refrigerated for storage and transport-adding to the cost and inaccessibility. In rural areas of Global South countries where snake bites are especially common, the treatment is particularly difficult to get. In contrast, the newly designed proteins are stable across a much wider range of temperatures, can potentially be produced in bulk using microorganisms like yeast, may prompt fewer side effects, and would be easier to fine tune and keep consistent. "These small de novo proteins have a number of really interesting advantages, including thermal stability, the cost of manufacturing, and the fact that they can target something in a way that an antibody might not be able to," Jardine explains. One day, such a product might be deliverable in an "EpiPen-like device," readily available out in the field where it's most needed, he suggests. Snake venoms are made up of many different toxins mixed together. Vázquez Torres and her colleagues focused their work on three-finger toxins (3FTx), deadly compounds that traditional antivenoms often perform poorly against. 3FTxs are especially prominent in the venom of elapids, the family of snakes which includes cobras, mambas, and coral snakes. These toxins (proteins themselves) wreak havoc in the mammalian body. Some are paralyzing neurotoxins, others destroy cells and damage tissue. The scientists sought to identify antidote proteins to combat three representative target toxins: a short-chain alpha neurotoxin, a long-chain alpha neurotoxin, and a cytotoxin. All three representative toxins are well studied, and so the scientists knew their intricate shapes from the start. From that base, they could identify the key binding sites they'd need to block to render each toxin inactive. They fed this information into the first of their AI tools called RoseTTAFold diffusion, a model similar to image generators like Dall-E and Midjourney, but one trained and specialized to output mock-ups of protein structures in accordance with requested criteria. In this case, the criteria were the toxin structures and the selected binding "hot spots," that the researchers were hoping to clog up. The AI offered up dozens of suggestions for neutralizing proteins (in the form of detailed images of protein configurations) that might fill those binding sites-like formulating keys for mystery locks. To understand more about these theoretical proteins and decode their makeup, Vázquez Torres, Baker and her co-authors deployed a second generative AI model called ProteinMPNN trained to produce feasible combinations of amino acids that could fold together to replicate the diffusion model's outputs. Protein folding is complicated and often hard to predict from amino acid sequences alone, and on the flipside, it's challenging to know what amino acid series will lead to which folded shapes. ProteinMPNN accelerates that computational process. Then, they used a third predictive AI tool called AlphaFold2 to independently predict how each of those amino acid strings would actually fold, thus double-checking the work of the prior two models. Between each step, the researchers applied their own expert human eyes to filter out duds and narrow the candidate pool to the best options. The study authors reverse-translated the most promising amino acid chains into DNA sequences, and then used modified bacteria to pump out the proteins. They tested their top candidates in a set of petri dish experiments with human muscle and skin cells, and found proteins effective against all three focal toxins. This narrowed the pool even further, down to one frontrunner per category. The scientists tested each of these in a series of mouse experiments. In initial tests, their anti-cytotoxin candidate didn't reduce skin lesions associated with envenomation, so the researchers ceased testing it. But the other two candidate proteins proved much more effective. When mixed directly with the target toxin, and injected into mice, both anti-neurotoxin proteins prevented all mouse deaths (without the added protective proteins, 100% of mice died). To mimic the process of treating a bite, the scientists then tested what happened to the mice when each toxin was administered first and the candidate proteins later. One of the proteins saved 100% of the mice it was given to, even administered up to 30 minutes after the toxin. The second protein prevented 80% of deaths administered after 15 minutes and 60% after half an hour. "It was shocking to see that these proteins work in animals, out of the box. We didn't need to do any optimization," says Vázquez Torres. "To find something that works on the first attempt, that's incredible." Moreover, the research went from idea to submitted publication data in just about a year, thanks to AI's computational assistance. "I think it's like record time for any kind of scientific paper," she says-demonstrating how much machine learning can accelerate the research process. The findings are just the latest in a recent wave of new developments in antivenom research, such as Jardine's synthetic antibodies and re-purposed pharmaceuticals. WHO designated snake envenomation a Neglected Tropical Disease in 2017, prioritizing snake bites for more investment and public health consideration. Since then, there's been a steady stream of studies. "This is adding another tool to the arsenal that we have to solve the problem. [The proteins] are going to have unique applications that antibodies don't and vice versa," Jardine says. Yet there's a long road ahead before de novo proteins can be approved for human use. The mouse trials didn't reveal any apparent negative side effects, though it's still unknown how these proteins act in the body and if they're truly safe. They're totally new molecules, and they'd need to be extensively screened and tested for off-target reactivity and adverse effects, note both Vázquez Torres and Jardine. "We need to prove these molecules are safe. We need to really understand their mechanisms," says Vázquez Torres. It will be years (and years) before any designer protein antivenom makes it to market. If it does, the proteins discovered by Vázquez Torres and her colleagues won't be enough. They only tackle two isolated toxins within certain venoms. Likely, around ten carefully designed proteins would need to be mixed together to neutralize a complete venom, says Vázquez Torres. In the hunt for a broad spectrum or universal antivenom, scientists are still searching. Still, the prospect of using microorganisms to pump out new-to-nature proteins on demand is thrilling to scientists. And the excitement goes beyond just antivenoms. De novo proteins could one day yield alternative therapies for all sorts of diseases. The amino acid constructions are somewhere between a biologic drug, made or derived from living organisms, and a small molecule drug like aspirin, which is chemically synthesized. "You can imagine a huge number of problems this could solve, that you couldn't solve with conventional approaches," Jardin says. "This is a really new way of doing things, and we're just scratching the surface."

New AI-based antitoxins achieve 100% survival against cobra venom

Baker's team, in collaboration with Timothy Patrick Jenkins from Denmark's Technical University (DTU), harnessed AI to design proteins that bind to and neutralize three-finger toxins -- among the deadliest components of cobra venom. These toxins are notorious for evading the immune system, rendering conventional treatments ineffective. In experiments, the new antitoxins achieved an 80-100% survival rate in mice exposed to lethal doses of three-finger toxins. While they do not yet neutralize the full complexity of snake venom, this success marks a significant step toward targeted, efficient treatments. "The antitoxins we've created are easy to discover using only computational methods. They're also cheap to produce and robust in laboratory tests," said Baker. The new proteins, produced using microbes, eliminate the need for animal immunization, potentially reducing production costs. Their small size could allow them to penetrate tissue more effectively and neutralize toxins faster than current treatments. Timothy Patrick Jenkins highlighted the broader potential: "The most remarkable result is the impressive neurotoxin protection they afforded to mice. However, one added benefit of these designed proteins is that they are small -- so small, in fact, that we expect them to penetrate tissue better and potentially neutralise the toxins faster than current antibodies. And because the proteins were created entirely on the computer using AI-powered software, we dramatically cut the time spent in the discovery phase. "