We’re living in a dangerous post-antibiotic era. AI can help.
In the time it takes you to read this article, one person in the US will die from an infection that antibiotics can no longer treat effectively.
And over the course of this year, 700,000 people around the world will die from drug-resistant infections. That annual death toll could rise to 10 million by 2050, a major UN report recently warned, if we don’t make a radical change.
Enter artificial intelligence.
For the first time, AI researchers have figured out how to identify brand-new types of antibiotics by training a neural network to predict which molecules will have bacteria-killing properties. They’ve just published their findings in the journal Cell.
The research team, based at MIT, has found a new compound that works on drug-resistant strains of M. tuberculosis, C. difficile, A. baumannii, and other pathogens when tested in mice. They named it halicin — after HAL, the AI system in 2001: A Space Odyssey — and used the word “excitingly” five times when describing their discovery in the study. It’s easy to see why: This is truly an exciting moment for both the AI community and the public health community, and it’s coming not a moment too soon.
The CDC warned in November that we’re now entering a post-antibiotic era — a time when our antibiotics are becoming pretty much useless. We’ve created this crisis by overusing antibiotics in the treatment of humans, animals, and crops. The bacteria have adapted to our drugs, morphing into superbugs that can all too easily decimate our health.
Big Pharma and biotech companies haven’t been creating new antibiotics because it takes many years and lots of funding to do the research and development. Most new compounds fail. Even when they succeed, the payoff is small: An antibiotic doesn’t sell as well as a drug that needs to be taken daily. So for many pharma companies, the financial incentive just isn’t there.
Looking at this deadlock, the MIT researchers thought: What if we could use AI to ramp up the speed of antibiotic discovery and drive down the cost? And that’s exactly what they did.
“I think it’s a breakthrough in a field of much unmet need,” said César de la Fuente, a bioengineer at the University of Pennsylvania who works on AI and antibiotics, and who was not involved in the MIT study. “After all, no new classes of antibiotics have been discovered for decades. This one is definitely structurally different from conventional antibiotics.”
Here’s how AI found a new type of antibiotic
AI excels at sifting through tons and tons of data, and antibiotic discovery requires just that. Scientists now have access to giant datasets in the form of chemical libraries, which catalog millions of known compounds. And while it may take humans years to search through so many candidates for that one miracle molecule, a neural network can do the work in days.
To start, the researchers behind the Cell study trained a neural network to identify molecules that fight E. coli bacteria by feeding it data on 2,335 molecules that we know have antibacterial properties. They then got the model to go through multiple chemical libraries containing a whopping 107 million molecules and predict which might fight E. coli effectively — while screening out the ones that resemble antibiotics we’ve already got. Finally, they took around 100 of the most promising hits and tested them physically in the lab.
The molecule they dubbed halicin turned out to be excellent at killing various bacteria, not only E. coli, when tested in mice. Best of all, the mice didn’t develop resistance to halicin, even after 30 days. (Resistance to other compounds sometimes develops within a day or two.) That’s crucial. There would be no point in developing a new drug only to have it, too, instantly fall prey to resistance.
The study shows how AI can help take the blinders off of scientists, who may get used to approaching a problem in a particular way. The neural network that found structurally new types of drugs did so without knowing any human-identified patterns in how various molecules tend to function. In other words, it had no preprogrammed assumptions, no limiting biases.
“As a result, the model can learn new patterns unknown to human experts,” a co-author of the study, Regina Barzilay, told Nature.
This isn’t the first time AI has shown promise in drug discovery. Just last month, a British startup called Exscientia claimed to have made the first AI-designed drug that will be clinically tested on humans. That drug is for obsessive-compulsive disorder.
Other researchers are using AI to hunt specifically for antibiotics but are using a different approach to the one favored by MIT. Rather than picking known molecules out of a database and seeing which are best at killing bacteria, Penn’s de la Fuente is using computers to actually design totally new molecules, unlike those we see in nature.
“The hypothesis I have is that perhaps the natural world has run out of inspiration,” de la Fuente said. “Perhaps we’ve found most of the interesting molecules that nature has produced with useful antibiotic properties, so it’s time to look elsewhere. Most likely, the next generation of antibiotics is not going to come from nature, but from machines.”
However, he added a note of caution. “AI is providing an exciting out-of-the-box approach to finding new antibiotics, but this is not going to solve the whole problem,” he said. We still need to stop the massive overuse of antibiotics that’s driving the drug resistance crisis.
As for halicin, the new compound discovered by the MIT team, the next step will be to test it in clinical trials. A lot of compounds that work in mice don’t work in humans, so for now we should limit our optimism to the cautious variety. Even if halicin does turn out to be highly effective in humans, it will be years before you’re able to get it as a shelf-ready antibiotic.
Still, this is a welcome and significant advance: Drug resistance is one of our worst public health nightmares, and AI is making impressive strides toward tackling it.
Sign up for the Future Perfect newsletter and we’ll send you a roundup of ideas and solutions for tackling the world’s biggest challenges — and how to get better at doing good.
Future Perfect is funded in part by individual contributions, grants, and sponsorships. Learn more here.
Author: Sigal Samuel
Read More
