Bacteria are evolving the ability to resist some of our most important antibiotics. Unless something is done, resistant strains of bacteria may kill 10 million people per year by 2050. Biomedical scientists are working hard on multiple approaches toward solving this problem.
In 2020, a team of American researchers reported a major breakthrough. For the first time, they indentified a new antibiotic using artifical intelligence.
They did this by using a deep learning neural network, which can learn to recognize patterns, from a training set of example of the pattern. The researchers trained the network with data on the molecular structures of 2,500 different drugs, and data about how effective each substance was as an antibiotic.
After training with the examples, the network was used to screen a library of 6,000 other substances, recognizing ones that both fit the learned pattern of a good antibiotic, and were different from existing antibiotics. Finding substances different from existing antibiotics was especially important, because it is likely that these will be harder for bacteria to evolve resistance to.
The researchers identified a promising new antibiotic which they named halicin, after the intelligent computer HAL-9000 from the sci-fi classic 2001: A Space Odyssey. Preliminary tests indicate that halicin has properties that will make it especially hard for bacteria to evolve resistance to it. After finding halicin, the researchers used their network to search a larger library of compounds and identified 23 more potential new antibiotics.
Reviewer: Thomas Unterthiner, Johannes Kepler University, Linz, Austia.
Sources and Further Reading
Y: About what, Don?
D: Bacteria are evolving the ability to resist some of our most important antibiotics. Unless something is done, resistant strains of bacteria may kill ten million people per year by 2050.
Y: Biomedical scientists are working hard on multiple approaches towards solving this problem. In 2020 a team of American researchers reported a major breakthrough. For the first time, they identified a new antibiotic using artificial intelligence.
D: How did they do that?
Y: They used a deep learning neural network. Neural networks can learn to recognize patterns, from a training set of examples of the pattern. The researchers trained the network with data on the molecular structures of two thousand five hundred different drugs, and data about how effective each substance was as an antibiotic. After training with the examples, the network was used to screen a library of six thousand other substances, recognizing ones that both fit the learned pattern of a good antibiotic, and were different from existing antibiotics. Finding substances different from existing antibiotics was especially important, because it is likely that these will be harder for bacteria to evolve resistance to.
D: I hope it worked.
Y: It did. The researchers identified a promising new antibiotic which they named halicin. Halicin was named after the intelligent computer HAL-9000 from the science fiction classic, Two Thousand One: A Space Odyssey. Preliminary tests indicate that halicin has properties that will make it especially hard for bacteria to evolve resistance to it. After finding halicin, the researchers used their network to search a larger library of compounds and identified twenty-three more potential new antibiotics.
