Antibiotic from the car
A powerful antibiotic has been found for the first time with the help of artificial intelligence
A team of scientists from the Massachusetts Institute of Technology reports that the new antibiotic halicin, found using machine learning technology, is able to cope with some of the most dangerous strains of pathogenic bacteria.
A computer model that can test more than a hundred million chemical compounds in a matter of days is designed to identify potential antibiotics that kill bacteria using mechanisms other than existing drugs. According to the creators of the project, this should ensure the effectiveness of the new antibiotic in relation to those bacteria that have developed resistance to other drugs. "We wanted to develop a platform that would allow us to use the capabilities of artificial intelligence to open a new era of antibiotic discovery," says Professor James Collins. "Our approach has revealed an amazing molecule that is arguably one of the most powerful antibiotics ever discovered."
Halicin was the first among antibiotics whose formula was discovered during computer screening of large digital libraries of chemical compounds. First, the researchers applied the deep learning method, providing the program with information about the atomic and molecular features of almost 2,500 drugs and natural compounds, as well as how well these substances blocked the growth of the E.coli bacterium. After the algorithm found out what molecular properties are inherent in good antibiotics, scientists passed through it data on more than six thousand chemical compounds. At the same time, the algorithm had to focus on finding substances that looked effective, but differed from existing antibiotics.
"The machine learning model can explore large numbers of molecular structures, which can be prohibitively expensive for traditional experimental approaches," says Professor Regina Barzilay. The idea of using predictive computer models for in silico screening is not new, but until now these models have not been accurate enough for drug discovery. Previously, the molecules in them were represented as vectors reflecting only the presence or absence of certain chemical groups.
The model selected one molecule that was predicted to have strong antibacterial activity, its chemical structure was different from any existing antibiotics. Using a different machine learning model, the researchers also showed that this molecule is likely to have low toxicity to human cells. This substance was previously considered as a potential drug for the treatment of diabetes.
E. coli colonies on which halicin (in the upper row) and ciprofloxacin (in the lower row) were tested.
The found substance was named halicin in honor of the HAL artificial intelligence system from the 2001 film Space Odyssey. After the molecular structure proposed by artificial intelligence was synthesized, it was tested on bacterial cultures and on laboratory mice. Among the bacteria that halicin was able to destroy were antibiotic-resistant Acinetobacter baumannii and Enterobacteriaceae, which are included in the list of bacteria recognized by the World Health Organization as particularly dangerous. In particular, the strain of Acinetobacter baumannii used in the study was resistant to all known antibiotics, but the use of ointment with halicin completely cured mice of infection in a day. Preliminary studies suggest that halicin kills bacteria by disrupting their ability to maintain an electrochemical gradient to move ions across cell membranes.
After detecting halicin, the researchers also used their model to screen more than 100 million molecules selected from the ZINC15 database. The screening, which took only three days, revealed 23 candidates that were structurally different from existing antibiotics and were predicted to be non-toxic to human cells. In laboratory tests against five types of bacteria, it was found that eight molecules showed antibacterial activity, and two were particularly strong. Now the researchers plan to conduct further testing of these molecules, as well as continue scanning substances from the ZINC15 database. They also plan to use their model to optimize existing antibiotics – for example, to train an algorithm to use it to make a certain antibiotic targeted only at specific bacteria, preventing it from destroying beneficial bacteria in the patient's digestive tract.
The study is described in the journal Cell (Stokes et al., A Deep Learning Approach to Antibiotic Discovery).
Portal "Eternal youth" http://vechnayamolodost.ru