Using machine learning algorithm to identify new antibiotic compounds, Using machine learning algorithms, MIT researchers have identified new effective antibiotics.

In laboratory tests, this drug has killed many of the most problematic bacteria in the world, including several strains that are resistant to all known antibiotics. It also eliminates infection in two different mouse models.

Computer models, which can filter out more than one hundred million compounds on a daily basis, were developed to select potential antibiotics that kill bacteria using different mechanisms from existing drugs.

We want to develop a platform that will enable us to use the power of artificial intelligence to usher in a new era of antibiotic discovery, researchers say. Our approach has found this extraordinary molecule, which is perhaps one of the most effective antibiotics found. Using machine learning algorithm to identify new antibiotic compounds.

In their new study, the researchers also identified several other promising antibiotic candidates who wanted them to further test. They believe that the model can also be used to develop new drugs based on knowledge of the chemical structure that allows drugs to kill bacteria.

Machine learning models can study large chemical spaces in silicon, which can be very expensive for conventional experimental approaches.

Barzillay and Collins, who teaches the faculty at the Abdul Latif Jamel Machine Learning Clinic at MIT Health, are the lead authors of this study, which appeared today on Cell.

Very few new antibiotics have been developed in recent decades, and most of the approved antibiotics are slightly different drug variants. Current screening methods for new antibiotics are often very expensive, require significant investment over time, and are usually limited to a narrow range of chemical diversity. Using machine learning algorithm to identify new antibiotic compounds.

The idea of ​​using predictive computer models for silico screening is not new, but so far this model has not been accurate enough to change drug discovery. Previously, molecules were presented as vectors that reflected the presence or absence of certain chemical groups.

New neural networks can learn this representation automatically by mapping molecules in continuous vectors, which are then used to predict their properties.

In this case, the researchers created their model to look for chemical properties that make the molecule effective in killing E. coli.

For this purpose, they trained models on about 2,500 molecules, including about 1,700 FDA-approved drugs, and on 800 natural products with diverse structures and diverse bioactivity. Using machine learning algorithm to identify new antibiotic compounds.

After the model was trained, the researchers tested it at the Broad Institute for Drug Reproduction, a library with around 6,000 compounds. The model selects molecules that are expected to have strong antibacterial activity and different chemical structures from all existing antibiotics.

Using other machine learning models, the researchers also showed that this molecule tends to have low toxicity to human cells.

This drug works on all species tested, except Pseudomonas aeruginosa, a pulmonary pathogen that is difficult to treat.

To test the effectiveness of chalicin in live animals, the researchers used it to treat mice infected with A. baumannii, a bacteria that infects many US troops stationed in Iraq and Afghanistan. Using machine learning algorithm to identify new antibiotic compounds.

The A. baumannii strain they use is resistant to all known antibiotics, but administration of an ointment containing halicin completely removes the infection within 24 hours.

Preliminary studies show that chalicin kills bacteria by affecting their ability to maintain electrochemical gradients in their cell membranes. This gradient is needed, for example, for the production of ATP (the molecule by which cells store energy) so that cells die when the gradient collapses.

This type of killing mechanism can make it difficult for bacteria to develop resistance, the researchers said.

In this study, the researchers found that E. coli did not develop resistance to chalicin during the 30-day treatment period.

In contrast, bacteria begin to develop antibiotic resistance to ciprofloxacin within one to three days, and after 30 days, the bacteria are about 200 times more resistant to ciprofloxacin than at the start of the experiment. Using machine learning algorithm to identify new antibiotic compounds.

The researchers plan to continue further studies on halicin by working with pharmaceutical companies or non-profit organizations to develop it for human use.

After identifying halicin, the researchers also used their model to filter more than 100 million selected molecules from the ZINC15 database, an online collection of about 1.5 billion compounds.

This screening, which lasted only three days, identified 23 candidates that were structurally different from existing antibiotics and were thought to be non-toxic to human cells.

The researchers also plan to use their model to develop new antibiotics and optimize existing molecules.

For example, they can train models to add features that make certain antibiotic targets only certain bacteria and prevent them from killing beneficial bacteria in the patient’s digestive tract.