X

AI can now design new antibiotics in a matter of days

IBM’s new AI system could also help design treatments for Covid-19

This story is part of a group of stories called ‘Finding the best ways to do good’.

Imagine you’re a scientist who needs to discover a new antibiotic to fight off a scary disease. How would you go about finding it?

Typically, you’d have to test lots and lots of different molecules in the lab until you find one that has the necessary bacteria-killing properties. You might find some contenders that are good at killing the bacteria only to realize that you can’t use them because they also prove toxic to humans. It’s a very long, very expensive, and probably very aggravating process.

But what if, instead, you could just type into your computer the properties you’re looking for and have your computer design the perfect molecule for you?

That’s the general approach IBM researchers are taking, using an AI system that can automatically generate the design of molecules for new antibiotics. In a new paper, published in Nature Biomedical Engineering, the researchers detail how they’ve already used it to quickly design two new antimicrobial peptides — small molecules that can kill bacteria — that are effective against a bunch of different pathogens in mice.

Normally, this molecule discovery process would take scientists years. The AI system did it in a matter of days.

That’s great news, because we urgently need faster ways to create new antibiotics.

Why antibiotic resistance is such a huge problem

When new antibiotics are introduced, they can have great, even lifesaving results. Since penicillin was discovered in 1928, kicking off the modern era of antibiotics, we’ve come to rely on them to treat killers like tuberculosis and to keep us safe when we undergo procedures like C-sections or joint replacements.

But experts have warned that we’re now entering a post-antibiotic era — a time when our existing antibiotics are becoming pretty much useless. We’ve created this crisis by overusing antibiotics in the treatment of crops, farm animals, and humans.

The more we overuse antibiotics, the more bacteria have a chance to adapt to our drugs, morphing into antibiotic-resistant superbugs that render our drugs ineffective.

And according to a new report from the Pew Charitable Trusts, the Covid-19 pandemic has aggravated the problem. Doctors have been even more inclined to unnecessarily prescribe antibiotics to patients. Even though Covid-19 is a viral illness and antibiotics don’t work on viruses, doctors have been giving patients these drugs to protect against secondary infections while they’re in the hospital — even before they know if the patients have infections or not.

The post-antibiotic era is here

Nowadays, 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 because of our antibiotic overuse. And over the course of the 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 warned, unless we make some radical changes.

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. For many pharmacompanies, the financial incentive just isn’t there.

But if you can use AI to do this work quickly and cheaply? Well, that just might change the calculus.

How IBM’s AI system works

IBM’s new AI system relies on something called a generative model. To understand it at its simplest level, we can break it down into three basic steps.

First, the researchers start with a massive database of known peptide molecules.

Then the AI pulls information from the database and analyzes the patterns to figure out the relationship between molecules and their properties. It might find that when a molecule has a certain structure or composition, it tends to perform a certain function. This allows it to “learn” the basic rules of molecule design.

Finally, researchers can tell the AI exactly what properties they want a new molecule to have. They can also input constraints (for example: low toxicity, please!). Using this info on desirable and undesirable traits, the AI then designs new molecules that satisfy the parameters. The researchers can pick the best one from among them and start testing on mice in a lab.

As one of the co-authors of the IBM paper, Aleksandra Mojsilović, told me, “You have the knobs to turn, and you get the molecule that satisfies the properties.”

The IBM researchers claim that their approach outperformed other leading methods for designing new antimicrobial peptides by 10 percent. They found that they were able to design two new antimicrobial peptides that are highly potent against diverse pathogens, including multidrug-resistant K. pneumoniae, a bacterium known for causing infections in hospital patients. Happily, the peptides had low toxicity when tested in mice, an important signal about their safety (though not everything that’s true for mice ends up being generalizable to humans).

Broader applications, from Covid-19 treatments to climate change solutions

This isn’t the first time AI has shown promise at solving longstanding problems in biology. Last year, the AI research lab DeepMind cracked the “protein folding problem” — the challenge of predicting which 3D shape a protein will fold up into — which has stumped biologists for 50 years and which has implications for drug discovery. Another exciting highlight: MIT researchers discovered a new type of antibiotic by training their AI to predict which molecules would have bacteria-killing properties.

The IBM research differs from MIT’s in an important way: Rather than training their AI on molecules that we know have antimicrobial properties (as MIT did), IBM trained theirs on a much broader database of all the known peptides that exist in nature. That’s the difference between starting out with around 100,000 data points and around 1.7 million data points.

The advantage of the latter is that you end up with an AI system that’s “more creative and generalizable,” according to Mojsilović. “We don’t want to be constrained to just antimicrobials. We really want to make a very generic tool that can be used in so many ways,” she told me.

Right now, for example, her team is working to figure out how the AI system might design treatments for Covid-19. When the pandemic came along, she explained, “We continued to say, we can use the same algorithms, but now we’re going to search a little differently — for something that looks like a molecule that can bind to a Covid target.”

In a blog post, the IBM researchers noted that while they’re excited about how the AI system can potentially accelerate antibiotic discovery and keep antibiotic-resistant bacteria at bay, they’re also hopeful that the system can have much broader applications. They envision it helping scientists “to discover and design better candidates for more effective drugs and therapies for diseases, materials to absorb and capture carbon to help fight climate change, materials for more intelligent energy production and storage, and much more.”

It’s not like this AI system will magically solve any of these problems on its own. But it advances a computational strategy for problem-solving that could yield really exciting benefits, and possibly save a lot of lives.

Siga Samuel