MIT’s Artificial Intelligence Platform Finds Antibiotics That Beyond Scientists’ Expectations

Humans have finally sacrificed super-artificial intelligence in a battle with superbugs. Not long ago, the James Collins team at the Center for Synthetic Biology at the Massachusetts Institute of Technology (MIT) Institute of Medical Engineering and Science, and the team led by Regina Barzilay of the German Institute, published major research in the top journal, Cell, in the form of a cover article.

MIT's Artificial Intelligence Platform Finds Antibiotics That Beyond Scientists' Expectations

James Collins (left) and Regina Barzilay (right)

They developed an artificial intelligence antibiotic prediction platform that jumps out of the human mind frame, which doesn’t need to know how drugs work, or even scientists to label chemical groups that can learn from one atom to another, and ultimately to predict specific molecules in a way that scientists don’t understand. Help humans find new antibiotics.

This new smart platform is so great that the Collins and Barzilay teams have used it to find a variety of antibacterial compounds that differ significantly from existing antibiotic structures. One of the preclinical drugs originally used to treat diabetes shows strong antibacterial effects and powerful lethality against a variety of super-resistant bacteria.

To make a difference, the antiseptic micromolecule kills bacteria in a way that scientists didn’t expect, and researchers say it’s hard for bacteria to become resistant to them.

MIT's Artificial Intelligence Platform Finds Antibiotics That Beyond Scientists' Expectations

Cell Cover

Since the birth of penicillin, antibiotics have become the cornerstone of modern medicine. With the widespread use of antibiotics, the problem of resistance of superbugs has become a major threat to human health. If the momentum of drug resistance is not curbed now, 10 million people are expected to die from drug-resistant bacterial infections by 2050.

Unfortunately, with traditional screening methods, it has become increasingly difficult to screen new antibiotics. The transformation of new antibiotics on the basis of existing antibiotics is also a failure.

In recent years, scientists looking for new antibiotics have begun to screen from large libraries of synthetic chemistry. Let’s not say that the operation of this method is difficult, high cost, the actual effect is actually very limited, from the beginning of 1980 to use this method, until now has not been applied to the clinical antimicrobial drugs appear.

Advances in machine learning technology and chemical informatics have given scientists another possibility: allowing artificial intelligence platforms to learn the characteristics of chemical molecules on their own and then predict its effects.

Collins and Barzilay’s team first developed an artificial intelligence platform that could learn the structure and characteristics of chemical molecules on its own, and then, based on the information it has, predicted the function of the molecules, which predicted whether they could inhibit the growth of specific bacteria. These chemicals are then sorted by the good or bad effect seof.

The researchers then fed the smart platform 2,335 different molecules, most of which were FDA-approved drugs and the rest were natural molecules that were widely bioactive.

After completing the training, the researchers asked the aia platform to identify compounds with antibacterial activity from 6,111 molecules in the Bode Institute compound library.

MIT's Artificial Intelligence Platform Finds Antibiotics That Beyond Scientists' Expectations

The screening process

Eventually, the platform found dozens of compounds with antibacterial potential. After a thorough analysis, a compound called SU332 came in first place. SU332 was originally c-Jun’s N-end kinase inhibitor, which is a preclinical treatment for diabetes.

Molecularly structurally, it is significantly different from any antibiotic available. In honor of the artificial intelligence system HAL 9000 in the classic sci-fi film “2001: Space Odyssey,” researchers named SU332 halicin.

Further analysis found that halicin killed bacteria rather than inhibited their growth and reproduction. Traditional penicillin kills e. coli in a metabolically inhibited state, and halicin can also kill. Even if the persistent bacteria left after antibiotic treatment, halicin can kill.

The researchers also used plasmids that carry a variety of drug-resistant genes to transform bacteria, giving them specific resistance, but these drug-resistant genes also failed to help bacteria resist halicin’s butcher’s knife.

The researchers then tested the knife with 36 multi-drug-resistant bacteria, and none of them escaped with the exception of copper-green pseudomonas. Subsequent analysis suggests that the copper-green pseudomonas can tolerate halicin, possibly because the “skin is too thick” , halicin can not get in.

MIT's Artificial Intelligence Platform Finds Antibiotics That Beyond Scientists' Expectations


So how on earth does halicin kill bacteria?

Preliminary research by the researchers suggests that the sterilization mechanism of halicin is not unusual. It does not destroy the structure of bacteria, but does not know how, destroy the ability of bacterial cell membranes to maintain electrochemical gradients. The electrochemical gradient of this cell membrane plays an important role in the synthesis of ATP.

ATP we all know that it is energy. If bacteria can’t produce ATPs, it’s going to die. The researchers believe that bacteria may have difficulty developing resistance to halicin’s bactericidal skills because it is difficult for bacteria to alter their ability to maintain electrochemical gradients through individual or mutations.

Subsequent studies of drug resistance confirm some of the researchers’ guesses.

They kept the bacteria in the lab for 30 days, and they were not able to obtain strains that were resistant to halicin. I don’t know if you remember, scientists at Harvard University and the Institute of Technology published a study in Science in 2016: E. coli is resistant to antibiotics that reach a minimum concentration of 1,000 times fatality in laboratory conditions, taking just 10 days.

MIT's Artificial Intelligence Platform Finds Antibiotics That Beyond Scientists' Expectations

The black background medium is divided into 5 concentration gradients (0-1000 times) from both sides to the middle, and the white bacteria on top of the medium are growing rapidly from both sides to the middle (antibiotic concentrations from low to high)

In subsequent studies, they tested halicin’s antibacterial effect on mice, which was really powerful. The research team is currently planning to work with pharmaceutical companies or public interest organizations to use halicin in humans.

To demonstrate the system’s powerful screening capabilities, researchers from Collins and Barzilay’s team then dropped more than 100 million chemical molecules into the system, from which they screened eight compounds with antibacterial activity, two of which were still very strong. The entire screening process took only 3 days.

I can’t imagine.

No wonder Roy Kishony, a professor of biology and computer science at the Color Institute of Technology, said, “This ground-breaking study marks a paradigm shift in the discovery of antibiotics, and even more generally. “