AI Is Now Creating Antidotes for Snake Venom | HackerNoon

Using artificial intelligence (AI) technology, a research team from the University of Washington found a novel way to produce snake venom antidotes. Their proof of concept could become a lifesaving solution that revolutionizes snakebite therapy.

Scientists Discovered AI Can Design Venom Antidotes

At the base of a serpent’s fangs sits a tiny sac filled with venomous liquid. When it bites, the venom travels through a narrow channel, out the tip and into the newly created wound. Within seconds, the wound site reddens and swells. The symptoms quickly worsen. Snakebite victims may soon experience blurred vision, nausea, cold sweats and difficulty breathing as the toxins work their way through the body.

Depending on the species, the symptoms can be even worse. Three-finger toxins (3FTxs) are potentially lethal components of Elapid venom. Serpents in the Elapidae family — cobras, mambas, kraits, death adders and coral snakes — are some of the most deadly. A bite from them can contain paralytic alpha neurotoxins (α-neurotoxins), which cause respiratory muscle failure. If left untreated, severe tissue damage and even death are likely outcomes.

Researchers recently discovered a way to treat such bites with the help of an AI solution inspired by generative technology. The discovery came from the laboratory of David Baker, a professor of biochemistry at the University of Washington.

Susana Vázquez Torres, a biochemistry student at the University of Washington and the paper’s lead author, believed her team could improve snakebite therapies with RFdiffusion — a guided diffusion model for developing new proteins.

Her belief was correct. Even though AI is still a developing technology, RFdiffusion was wildly successful. The team used it to design small proteins that bind strongly to the target proteins, counteracting snake venom. This solution neutralizes toxins, helping snakebite victims survive with virtually no early adverse reactions (EARs).

How Scientists Used AI to Neutralize Lethal Snakebites

The team’s deep learning technology created protein designs that matched the shape of the molecules they wanted to target. Instead of focusing on the binding site conventional antibody therapies do, they used AI to create high-affinity binders that focus on beta strands (β-strands) — a linear arrangement of amino acids.

α-neurotoxins have a multistranded beta structure, meaning they consist of β-strands. These short-chain and long-chain toxins bind to different proteins. Since many Elapid serpents derive their lethality from these toxins, neutralizing them is critical. This is why the researchers focused on β-strands instead of on the conventional binding site.

In short, the deep-learning-based solution created antivenoms for Elapid toxins. RFdiffusion trajectories guided generation toward edge β-strands in each neurotoxin, reducing noise one strand at a time. It effectively recognized key regions of α-neurotoxins and generated small proteins that matched the shape of the target proteins.

The designs had high binding affinity, exceptional thermal stability and a near-atomic-level agreement with the computational models. They neutralized 3FTx in vitro, so the researchers progressed to testing within the body to mimic real-life snakebite scenarios.

Of the 78 designs that the RFdiffusion produced, the best was SHRT. The biochemistry students injected multiple groups of laboratory mice with a mixture of binder and toxin and observed the effects. Whether SHRT was administered 15 minutes or 30 minutes after a lethal dose of α-neurotoxin was introduced, the mice survived.

Thanks to RFdiffusion, the new protein achieved 100% protection when administered at a 1:10 molar ratio — the ratio between the amounts in moles in the binder and toxin — and at a 1:5 molar ratio. Interestingly, SHRT also provided 100% protection against the short-chain consensus α-neurotoxin, which is the name for the most dangerous Elapid toxins.

Is This Method Better Than Antibody-Based Treatment?

Snakebites are often deadly. According to the World Health Organization, snakes bite approximately 5.4 million people each year. Out of the estimated 2.7 million injected with venom, as many as 137,880 die annually. Another 413,640 undergo amputations or develop disabilities as a result of the toxins.

Today, the only treatment for snakebites involves antibodies sourced from the plasma of immunized animals like sheep or horses. Trained professionals have to extract venom manually, inject it into the animal and then harvest plasma. This time-consuming process is challenging for several reasons.

For one, antivenoms based on animal-derived antibodies are relatively ineffective against toxins that don’t provoke a significant immune response, making them subpar for anything in the 3FTx class. Even if their efficacy was exceptional, producing them is complicated. Due to the nature of antibody development, inherent batch-to-batch variations exist.

What’s worse, traditional antibody-based treatment tends to cause severe EARS. The incidence rate is as high as 37.8% in some cases. Within 24 hours, patients experience symptoms ranging from nausea to anaphylactic shock. Conversely, the mice injected with the binder alone experienced no ill effects.

In addition to being easier to produce and safer to use, the strategy involving RFdiffusion and SHRT is faster and cheaper. Elapid snakes can often be found in regions within Africa and Latin America — they prefer tropical and subtropical regions. In areas with limited resources, an affordable, effective solution like this one could be revolutionary.

The Implications of AI-Generated Snake Venom Antidotes

The safe, stable toxin-neutralizing proteins developed by the University of Washington’s research team provide an accessible, cost-effective alternative to antibody-based treatment for snakebite victims. They might have discovered the next generation of antivenom treatment.

Even when accounting for the cost of developing, deploying and maintaining AI, this solution may still be cheaper than traditional therapeutics. Since this is a proof of concept, there is no evidence for this claim yet. However, other applications of AI demonstrate its cost-saving effects. According to Harvard’s School of Public Health, leveraging AI in diagnostics can reduce health costs by as much as 50% while improving health care by 40%.

Crucially, this deep-learning-based solution doesn’t just help snakebite victims — it could revolutionize multiple domains in health care, potentially helping millions of individuals. RFdiffusion could democratize research into autoimmune diseases or cancers, accelerating research and making treatments more affordable.

This Deep-Learning-Based Technology Could Save Lives

So far, targeting toxins in the 3FTx class with a deep-learning-based model for developing new proteins is an emerging concept. The health care sector is slow to change, so it may not be used in practice for a long time. However, if decision-makers recognize the promise of AI, this solution could save lives within just a few years.