On February 20th, local time, the Massachusetts Institute of Technology (MIT) published a breakthrough in the international academic journal Cell Sexual research results, and appeared on the cover of the current issue.

Researchers have used artificial intelligence deep learning systems to discover a powerful new antibiotic compound. In laboratory tests, the drug killed many of the world’s most difficult pathogenic bacteria, including some that are resistant to all known antibiotics. In animal experiments, it also effectively cleared bacterial infections in two different mouse models.

“We want to develop a platform that lets us use artificial intelligence to start a new era of antibiotic drug discovery “. James J. Collins, corresponding paper author and professor of the Department of Medical Engineering and Science at the Massachusetts Institute of Technology “Our method revealed this magical molecule, it may be one of the most powerful antibiotics found so far .”

A model built by researchers using computer deep learning systems, can screen more than 100 million compounds in a few days to select potential antibiotics that are different from existing drug-killing mechanisms.

In addition to discovering new antibiotics that can target multiple super-resistant bacteria, researchers have also identified several other promising candidate antibiotics that they plan to test further. Researchers believe that this model can also be used to design new drugs that, based on their knowledge of the chemical structure, enable the drugs to kill bacteria.

A new era of drug discovery

Traditional methods for screening new antibiotics are costly, not only require a lot of time investment, but are usually limited to a narrow range of chemical diversity, In the past few decades, few new antibiotics have been developed, and most newly approved antibiotics are slightly different variants of existing drugs.

“ We are facing a growing crisis in antibiotic resistance, a situation in which more and more pathogens are becoming resistant to existing antibiotics, and the biotechnology and pharmaceutical industries are Caused by a shortage of antibiotics. “Collins said.

Traditional drug screening models are not precise enough. Molecules are expressed as carriers that reflect the presence or absence of certain chemical groups and cannot change the way drugs are discovered. However, new neural networks can automatically learn these representations, map molecules into continuous vectors, and then use them to predict their properties.

James J. Collins academic big data, scan the QR code in the figure for details (Source: AMiner)

Collins has previously developed machine learning computer models that can be trained to analyze the molecular structure of compounds and correlate them with specific features.

Based on this, the researchers designed a new model looking for chemical characteristics that enable the molecule to effectively kill E. coli. To do this, they trained about 2,500 molecules on the model, including about 1,700 FDA-approved drugs and 800 natural products with different structures and broad biological activities.

Machine learning model for antibiotic screening

Once this model has been trained, researchers are at the MIT and Harvard Broad Institute’s Center for Drug Reuse (Drug repurhub) Test them.

Finally, The model singled out a molecule predicted to have strong antibacterial activity, with a chemical structure different from any existing antibiotic. The researchers also discovered through another machine learning model that this molecule may have low toxicity to human cells.

According to the fictional artificial intelligence system in (2001: A Space Odyssey) in 2001, the researchers decided to name this molecule Halicin.

The researchers then tested dozens of strains isolated from patients. By culturing in a laboratory dish, the researchers found that the potential drug could kill many bacteria resistant to treatment, including Clostridium difficile, Acinetobacter baumannii, and Mycobacterium tuberculosis. Except for a Pseudomonas aeruginosa (Pseudomonas aeruginosa) This potential drug is tested for All drug-resistant bacteria have shown significant effects.

To test the effectiveness of halicin in live animals, researchers used it to treat mice infected with Baumannii, a bacteria that has previously infected many American soldiers stationed in Iraq and Afghanistan, and Bowman Nisseria is resistant to all known antibiotics, but after treatment with halicin, the bacterial infection completely disappeared within 24 hours.

Halicin shows significant effect in mouse infection model

Further research shows that the new drug halicin kills bacteria by disrupting their ability to maintain an electrochemical gradient on the cell membrane. Among them, the electrochemical gradient is necessary to generate ATP (the molecule used by the cell to store energy) , so if the gradient is broken, the cell will die . Researchers also mentioned that the process of reshaping the electrochemical gradient is very complicated and can not be completed by just a few mutations, so this also prevents the development of drug resistance to the greatest extent.

“ When you deal with a molecule that may be related to a membrane component, a cell does not necessarily acquire one or two mutations to change the chemical properties of the outer membrane. From an evolutionary perspective, such mutations require Much more complicated, “said Jonathan Stokes, the first author of the paper and postdoctoral fellow at MIT and the Harvard Broad Institute.

In addition, the researchers found in this study that E. coli did not develop any resistance during the 30-day treatment period. By contrast, the bacteria started to develop resistance to the antibiotic ciprofloxacin within 1-3 days. After 30 days, the bacteria were about 200 times more resistant to ciprofloxacin than at the beginning of the experiment.

Researchers also plan to work with pharmaceutical companies or non-profit organizations to further develop HalicinFurther research to develop drugs for humans.

Discover more potential drugs

After identifying Halicin, the researchers also used their model in the ZINC15 database (ZINC15 is an online collection of about 1.5 billion Database of compounds) More than 100 million molecules were screened.

In just three days, this screening identified 23 candidate antibiotic drugs that are structurally different from existing antibiotics and are non-toxic to human cells.

In laboratory tests on five bacteria, researchers found that eight of the candidate antibiotic molecules showed antibacterial activity, two of which were particularly strong. Researchers now plan to further test these molecules while screening more compounds in the ZINC15 database.

Predicting new antibiotic candidates from the massive chemical library

Researchers also plan to use their models to design entirely new antibiotics and optimize existing molecules. For example, they can train models to add properties that target specific antibiotics to specific bacteria and prevent it from killing beneficial bacteria in the patient’s digestive tract.

Roy Kishoni, professor of biology and computer science at the Israel Institute of Technology (Roy Kishony) said, This breakthrough The research is a landmark change in the development of antibiotic drugs, which is expected to improve our efficiency in discovering new antibiotics and bring us more weapons against super bacteria.

Reference:

http://news.mit.edu/2020/artificial-intelligence-identifies-new-antibiotic-0220

Jonathan M. Stokes et al., (2020), A Deep Learning Approach to Antibiotic Discovery, Cell, DOI: https://doi.org/10.1016/j.cell .2020.01.021