Researchers have revealed that artificial intelligence (AI) has invented two new antibiotics that could potentially kill drug-resistant gonorrhea and methicillin-resistant Staphylococcus aureus (MRSA).
Designed atom by atom, these drugs killed these superbugs in laboratory and animal tests.
Both compounds still require years of refinement and clinical trials before they can be used.
But the team behind them, from the Massachusetts Institute of Technology (MIT), says AI could usher in a “second golden age” of antibiotic research and development.
Antibiotics kill bacteria, but drug-resistant infections currently kill over a million people each year.
Overuse of antibiotics has led bacteria to evolve the ability to evade drugs, and new antibiotics have been in short supply for decades.
Researchers have previously used artificial intelligence to screen thousands of known chemicals, hoping to identify potential new antibiotics.
Now, a research team at MIT has gone a step further, using generative artificial intelligence (GPUs) to design antibiotics for the sexually transmitted infection gonorrhea and the potentially deadly Staphylococcus aureus.
Their study, published in the journal Cell, reports that they have examined 36 million compounds, including those that do not exist or have yet to be discovered.
The scientists trained the AI by feeding it data on the chemical structures of known compounds and whether they slow the growth of different types of bacteria.
The AI then learned how bacteria are affected by different molecular structures made of atoms such as carbon, oxygen, hydrogen, and nitrogen.
The researchers then tried two approaches to using AI to design new antibiotics. The first method involves searching a library of millions of chemical fragments (ranging from 8 to 19 atoms in size) to find a promising starting point and build upon it. The second approach is to give AI free rein from the outset.
The design process also eliminates any ingredients that are too similar to existing antibiotics. It also strives to ensure they are creating medicines, not soaps, and filters out any substances expected to be toxic to humans.
Scientists have used AI to develop antibiotics for gonorrhea and Staphylococcus aureus, a bacterium that is harmless on the skin but can cause serious infections once it enters the body.
After production, these leading designs were first tested on mice, ultimately leading to the development of two new potential drugs.
Professor James Collins, researcher at MIT
“We’re very excited because we’ve demonstrated that generative AI can be used to design entirely new antibiotics,” Professor James Collins of MIT told the BBC.
“AI can allow us to rapidly and cheaply develop molecules that will expand our arsenal and truly give us an edge in the battle of wits against superbugs.”
However, these drugs are not yet ready for clinical trials and require further refinement before the lengthy process of human trials, expected to take another year or two.
Dr. Andrew Edwards of the Fleming Initiative and Imperial College London said the work was “significant” and had great potential because it “demonstrates a new way to identify novel antibiotics.”
But, he said, “while AI has the potential to significantly improve drug discovery and development, we still have a lot of hard work ahead of us in terms of safety and efficacy testing.”
This can be a long and expensive process, and there’s no guarantee that experimental drugs will ultimately be used in patients.
Some have called for more iterative approaches to AI-based drug discovery.
“We need better models,” Professor Collins said. These models should not only focus on how drugs perform in the lab but also better predict their effectiveness in the body, he said.
The challenges of manufacturing AI-designed drugs are also a concern. Of the 80 best theoretical gonorrhea treatments designed, only two have been synthesized for production.
Professor Chris Dowson of the University of Warwick said the research was “very cool” and showed that AI “is a significant step forward as a tool in antibiotic development to mitigate the emergence of drug resistance.”
However, he explained, drug-resistant infections also present an economic problem: “How do you make a commercially viable drug?”
If a new antibiotic is invented, ideally, people would use it as little as possible to maintain its effectiveness, making it difficult for anyone to profit from it, he said.