Proteins designed by artificial intelligence have protected mice against lethal snake toxins for the first time, in a "breakthrough" moment that could turbocharge the race for safer, more effective antivenoms.
In a project led by Prof David Baker, who this year shared the Nobel Prize in Chemistry, researchers used machine models to create the new proteins.
Also known as "binders", these proteins are the building-blocks of antivenom treatments. They are designed to attach to dangerous toxins, effectively disarming them.
"For the first time globally, we showed that these AI-designed binders also work in living creatures," Dr Timothy Jenkins, an associate professor at the Centre for Antibody Technologies in Denmark and co-author of the paper, told the Telegraph.
"In different conditions, we were able to save 100 per cent of mice against neurotoxicity from snake venom toxins," he said, adding that "the same tech can work for viruses, cancer, or autoimmune disease."
Researchers say the study, published in Nature on January 15, is exciting because it shows how the AI-driven design of proteins can dramatically accelerate the development of new therapies.
"We started this collaboration a bit over a year ago, and within a couple of weeks we had very, very strong binders against [snake toxins] that we'd struggled with for two years in the lab," said Dr Jenkins.
"It is a game changer, because [AI] makes the discovery of therapeutics way more cost effective and a lot faster."
In the snakebite field, this could prove critical. Each year, between 81,000 and 138,000 people die from snakebites, the vast majority in poor rural regions of the developing world, while a further 400,000 suffer life changing injuries including amputations, sight loss and open ulcers that never heal.
Yet treatment options have barely changed for a century. Not only is the production of traditional antivenoms laborious (it involves milking snakes, before injecting horses with toxins and harvesting their antibodies) but the resulting antivenoms are difficult to distribute and administer.
Side effects can be so severe that existing antivenoms can only be administered in medical facilities, where rehabilitation from anaphylactic shock is possible.
Dr Diogo Martins, head of the snakebite programme at the Wellcome Trust and not involved in the research, told the Telegraph he is cautiously optimistic about the study.
"While I wouldn't describe it as revolutionary - there are several groups working on similar approaches - it does stand out for effectively translating machine learning data into actual efficacy with thermostable, production-friendly proteins. That's a significant step forward and could address some of the persistent challenges in this space.
"That said, as with all such approaches, the cost of goods will be critical," he added. "Scaling this technology for broader application, particularly in resource-limited settings where the need is greatest, remains a key hurdle."
To date, updating antivenoms has proved a notoriously difficult task - partly because snake venoms have evolved over centuries to become some of the most complex toxins on earth, which means they contain many targets which binders need to neutralise.
But researchers are excited about the possibilities of rapidly designing the proteins that form the foundations of any snake antivenom, without milking snakes or injecting horses.
"Typically when we're making antibodies, like in the case of Covid-19, you've got one target - you make the best antibody, and you put all of your eggs into one basket," said Dr Jenkins.
"But here, we need to cover this huge target space of about 2,000 toxins, if we want to cover snakebite as a whole. Even if we take a single species, it can add up to 100 different toxins, so it's not a very simple thing to solve, even if you just want to neutralise venom from a single snake."
Finding proteins known as "broadly neutralising binders", which work against a broad range of snake toxins, is becoming more feasible because of the breakthrough in AI.
Prof Baker's Nobel-prize was shared with Demis Hassabis and John Jumper of Google's AI company DeepMind. Prof Baker created entirely new kinds of proteins, while Hassabis and Jumper developed an AI model called AlphaFold2 which has been able to predict the structure of virtually all the 200 million existing proteins that have been identified.
The use of these and other AI technologies to develop antivenoms is causing excitement, not least because the new proteins appear to be very stable, meaning they can easily be mass-manufactured and distributed without the need for a cold chain. The researchers mainly used three programmes, called Alphafold 2, RFdiffusion and ProteinMPNN.
"In terms of next steps ... we actually want to make a [human] product now," said Dr Jenkins.
"We've conducted a proof of concept study showing that this tech works.... I think six months isn't an unrealistic timeframe to have a panel of therapeutic binders for a specific geographical region to test [in a laboratory]."