A machine-learning algorithm has been developed to help find new antibiotics and has succeeded in identifying a molecule that has powerful antibiotic properties and can kill several types of bacteria that cause disease in humans, including strains that are resistant to most other types of antibiotics currently in use.
Resistance to antibiotics is a growing problem. Several strains of bacteria have now acquired resistance to virtually all types of antibiotics currently approved for use. The situation is likely to get far worse unless totally new antibiotics can be found. Finding potential drugs to treat microbial infections is a time consuming and expensive process, and with fewer economic incentives to search for new antibiotics, private sector drug discovery fallen in recent years. According to a review of antimicrobial resistance by J. O’Neill in 2014, by 2050 there will be 10 million deaths per year due to microbes that are resistant to antibiotics.
The AI system was developed by a team of microbiologists and computer scientists at MIT who trained the system to predict molecules that would have antimicrobial properties. They used the system to analyze millions of molecules from a variety of chemical libraries. The system analyzed the entire data set in a few days, eliminating the need for many thousands of experiments. The researchers feed the libraries into the AI system and then only have to test the molecules the system predicts will have antimicrobial properties.
The system found one molecule in the Drug Repurposing Hub that the researchers named Halicin that was structurally different from conventional antibiotics and had strong antimicrobial properties against pathogens from a broad phylogenetic spectrum. The tests showed Halicin was effective against Mycobacterium tuberculosis, Escherichia Coli, carbapenem-resistant Enterobacteriaceae, Clostridioides difficile, and pan-resistant Acinetobacter baumannii infections in murine models. The molecule works by disrupting the electrochemical gradient across bacterial cell membranes.
“Our approach revealed this amazing molecule which is arguably one of the more powerful antibiotics that has been discovered,” said lead researcher, Professor James Collins.
The researchers also used 23 empirically tested predictions to analyze more than 107 million molecules in the ZINC15 database and identified 8 molecules that were structurally different from currently used antibiotics that have been predicted to have strong antimicrobial properties.
You can read more about the research in the paper – A Deep Learning Approach to Antibiotic Discovery – which was recently published in the journal Cell. DOI: 10.1016/j.cell.2020.01.021