Scientists Used AI To Discover a New Powerful Antibiotic
It’s been recently published in the scientific journal Cell that a powerful new type of antibiotics has been discovered, using a pioneering machine-learning method. The work was led by biologist Jim Collins at the Massachusetts Institute of Technology in Cambridge.
The definition of machine-learning is the application of artificial intelligence (AI) to learn from its own experience. This is the ability of computerized systems to learn and improve from ‘their own’ experience without being explicitly programmed, like playing chess against itself without human players.
This is the first time that the use of AI has led to the identification of a completely new kind of antibiotic from scratch, without the input of any previous human assumptions. The new antibiotic was named halicin. When tested on mice, it was found to be highly effective against a wide range of bacteria, including tuberculosis and strains that were considered ‘pan-resistant’ and thus untreatable.
In previous experiments involving other antibiotic compounds, resistance has typically arisen within a day or two, but in the case of halicin it did not occur even after 30 days. The name halicin was picked as a homage to HAL, the intelligent computer in the movie 2001: A Space Odyssey.
Why Is the Discovery So Significant?
In recent years bacterial resistance to antibiotics has bee rising exponentially, while the discovery and regularity approval of new antibiotics is slowing down. Experts have made the grim prediction that by the year 2050, infections could become resistant to the point where they would kill 10 million people per year. One of the ways to avoid this terrifying scenario is by finding new antibiotics. The problem, according to Collins, is that people keep finding the same molecules again and again. That’s why the development of new modes of searching is so important.
How Was AI Used to Discover the New Antibiotic?
The team of researchers developed an AI algorithm inspired by the brain’s structure. This system, called a neural network, learns the features and qualities of molecules atom by atom. The neural network was then trained to spot molecules which hinder the growth of the bacterium known as E. coli.
Once the system was trained, it was used to screen about 6000 molecules. The researchers asked the model to predict which ones would be effective against E. Coli and to show only the ones that look different from conventional antibiotics. Out of the results, about 100 candidate molecules were selected for physical testing. Luckily, one of these molecules - halicin - turned out to be an extremely potent antibiotic.
The algorithm predicts the functions without any assumptions about how drugs work and without chemical groups being labeled, and it’s because of that neutral approach it can learn new patterns, unknown to human experts.
What Does It All Mean for The Future?
AI has already been put to use for medical research purposes, but in a slightly different way than Collins and his team have used it here. Instead of searching for specific structures and molecular classes, their network looked for molecules that display a particular activity.
The team hopes to broaden the general approach for finding new antibiotics and even use their methods to design molecules from scratch. The use of AI technology within healthcare is still considered in its infancy, but it is proving a powerful tool that may help us reach some major breakthroughs in the medical field. Many experts believe the next level of medicine and medical drugs depends enormously on our progress in computer processing power.
The definition of machine-learning is the application of artificial intelligence (AI) to learn from its own experience. This is the ability of computerized systems to learn and improve from ‘their own’ experience without being explicitly programmed, like playing chess against itself without human players.
This is the first time that the use of AI has led to the identification of a completely new kind of antibiotic from scratch, without the input of any previous human assumptions. The new antibiotic was named halicin. When tested on mice, it was found to be highly effective against a wide range of bacteria, including tuberculosis and strains that were considered ‘pan-resistant’ and thus untreatable.
In previous experiments involving other antibiotic compounds, resistance has typically arisen within a day or two, but in the case of halicin it did not occur even after 30 days. The name halicin was picked as a homage to HAL, the intelligent computer in the movie 2001: A Space Odyssey.
Why Is the Discovery So Significant?
In recent years bacterial resistance to antibiotics has bee rising exponentially, while the discovery and regularity approval of new antibiotics is slowing down. Experts have made the grim prediction that by the year 2050, infections could become resistant to the point where they would kill 10 million people per year. One of the ways to avoid this terrifying scenario is by finding new antibiotics. The problem, according to Collins, is that people keep finding the same molecules again and again. That’s why the development of new modes of searching is so important.
How Was AI Used to Discover the New Antibiotic?
The team of researchers developed an AI algorithm inspired by the brain’s structure. This system, called a neural network, learns the features and qualities of molecules atom by atom. The neural network was then trained to spot molecules which hinder the growth of the bacterium known as E. coli.
Once the system was trained, it was used to screen about 6000 molecules. The researchers asked the model to predict which ones would be effective against E. Coli and to show only the ones that look different from conventional antibiotics. Out of the results, about 100 candidate molecules were selected for physical testing. Luckily, one of these molecules - halicin - turned out to be an extremely potent antibiotic.
The algorithm predicts the functions without any assumptions about how drugs work and without chemical groups being labeled, and it’s because of that neutral approach it can learn new patterns, unknown to human experts.
What Does It All Mean for The Future?
AI has already been put to use for medical research purposes, but in a slightly different way than Collins and his team have used it here. Instead of searching for specific structures and molecular classes, their network looked for molecules that display a particular activity.
The team hopes to broaden the general approach for finding new antibiotics and even use their methods to design molecules from scratch. The use of AI technology within healthcare is still considered in its infancy, but it is proving a powerful tool that may help us reach some major breakthroughs in the medical field. Many experts believe the next level of medicine and medical drugs depends enormously on our progress in computer processing power.