Artificial intelligence finds new antibiotics to kill super-resistant bacteria

Today, the leading academic journal, Cell, publishes a research paper from the Massachusetts Institute of Technology (MIT). Scientists have discovered the antibacterial potential of a potential diabetes drug through a deep learning system that allows artificial intelligence to “see the eye.” In animal experiments, the new antibiotic is effective in killing a superbug that is resistant to all known antibiotics. The blockbuster discovery also appeared on the cover of Cell magazine.

Artificial intelligence finds new antibiotics to kill super-resistant bacteria

Image credit: Amanda Cicero, Luca Vallescura, Darryl “Moose” Norris, and Chris Sinclair

How did scientists come up with artificial intelligence to find new antibiotics? At the beginning of the paper, they say that since the birth of penicillin, antibiotics have become one of the cornerstones of modern medicine. However, with the abuse of antibiotics, more and more bacteria are resistant to antibiotics. Unfortunately, many antibiotics used to come from microbes in the soil, and it was not easy to develop them in the development of traditional drugs. It is not hard to see why, in the past few decades, few new antibiotics have been born, and they are structurally similar to those that have existed in the past.

To change this dilemma, the researchers developed a machine learning model. Specifically, this model can automatically learn the structure of different drug molecules, not only to know whether there are specific chemical groups at different locations of these molecules, but also to predict the properties of these molecules.

Artificial intelligence finds new antibiotics to kill super-resistant bacteria

Illustration of this study (Image Source: Resources)

The researchers then provided the model with 2,335 different molecules for “learning,” including drugs approved by the FDA, and many natural molecules with wide biological activity. The researchers hope that after training, the model will learn to identify drugs that kill E. coli.

After training, it’s time to test the learning abilities of this machine learning model. Using a library of compounds from the Broad Institute, the researchers asked the model to look for molecules with potential antibacterial potential from 6,111 of them. From this, this model suggests that a molecule has strong antibacterial activity. Interestingly, this molecule was originally developed as a diabetes drug and is structurally different from any antibiotic available. Subsequent studies have also shown that the molecule is less toxic to human cells.

According to MIT news, researchers paid tribute to the classic sci-fi film “2001 Space Odyssey” (2001: A Space Odyssey), which named the molecule halicin (the artificial intelligence system in the film called HAL 9000). They then tested halicin’s bactericidal effect on a variety of resistant bacteria in a petri dish, and the results were pleasing! With the exception of Pseudomonas aeruginosa, a difficult lung pathogen, halicin is killing all drug-resistant bacteria tested.

Artificial intelligence finds new antibiotics to kill super-resistant bacteria

  In addition to copper-green pseudomonas (blue), halicin showed good broad-spectrum antibacterial activity in several drug-resistant bacteria tested (Photo: Resources 1)

Of course, the results in a petri dish do not yet represent the antibacterial efficacy of living animals. The researchers then infected the mice with a super-resistant bacillus( A. baumannii). Also according to MIT news, this superbug can withstand all known antibiotics! And halicin is once again showing a magical effect – the ointment containing halicin completely clears the infection within 24 hours.

Based on these results, the researchers point out that halicin has broad-spectrum antibacterial activity. At the point of view, this is because it interferes with bacteria and does not allow them to form electrochemical gradients across membranes. In general, this electrochemical gradient helps bacteria generate energy. Without this gradient, the bacteria would die. The researchers also note that the process of reshaping electrochemical gradients is complex and can be done without a few simple mutations, so this also maximizes the elimination of drug resistance.

Using the system, the researchers further screened hundreds of millions of molecules in another database and found 23 potential antibacterial molecules that are very different from existing antibiotic structures and are not toxic to human cells. This screening process took only a short period of 3 days. Subsequent studies also showed that eight of these molecules did have anti-bacterial activity, and two molecules were particularly active. Scientists also plan to continue to study and evaluate these molecules.

As some scientists have said, this ground-breaking study is a paradigm shift in antibiotic drug development that promises to make new antibiotics more efficient and give us more weapons to fight superbugs.