- Using artificial intelligence, researchers found a powerful antibiotic named halicin.
- It kills harmful bacteria, including ones that are resistant to all existing antibiotics.
- It has a different chemical structure from all known antibiotics and uses a different approach to kill bacteria.
The discovery of penicillin (the first antibiotic) revolutionized medicine in the 20th century. However, the development of antibiotics has been significantly reduced over the past few decades.
Antibiotics discovered in recent years are just slightly different versions of existing drugs. Biologists have put great amounts of effort screening soil-dwelling microbes for features known to kill deadly bacteria.
But since superbugs resistant to modern antibiotics have spread widely, scientists haven’t been able to achieve major breakthroughs. Moreover, the existing techniques for screening new antibiotics require substantial time and money investment and are often limited to chemical diversity.
Now, artificial intelligence is helping biologists to look into large databases of molecules that are very different from existing antibiotics. Recently, MIT researchers used a novel machine-learning algorithm to analyze a massive digital archive of over 100 million chemical compounds and identify those that can kill many of the world’s most problematic disease-causing bacteria.
They have discovered a new powerful antibiotic compound that kills harmful bacteria, including ones that are resistant to all existing antibiotics.
They used a deep learning model (an increasingly popular subset of machine learning method) to detect antibiotics that use different approaches from existing drugs to kill bacteria.
In addition to this new antibiotic, the model was able to identify 23 other promising antibiotics candidates that had different structures as compared to existing drugs and predicted to be non-toxic to human cells.
The model can analyze large chemical spaces that are usually time-consuming and costly to explore with conventional experimental methods. According to the research team, the model has learned a lot about chemical structure in its training phase, so it could also be utilized in designing new drugs.
Halicin: A Powerful Antibiotic
The model maps molecules into continuous vectors, which reflect the absence or presence of specific chemical groups. These vectors are subsequently used to estimate the properties of compounds.
The team trained their model on nearly 2,500 molecules to search for properties that make molecules effective at killing E.coli. Once trained, they tested the model on 6,000 compounds from the Broad Institue’s Drug Repurposing Hub.
The trained neural networks detected one molecule that was estimated to have very effective antibacterial properties, low toxicity towards human cells, and a chemical structure dissimilar to all existing antibiotics.
Halicin (top row) vs. ciprofloxacin (bottom row) | Courtesy of Collins Lab at MIT
Researchers have named it halicin, after the AI system from a sci-fi movie ‘2001: A Space Odyssey’. They tested this molecule against several lab-grown bacterial strains. It worked against most of the species that they tested, including bacteria that became resistant to all known antibiotics.
The team also tested halicin in two different mouse models. It was able to completely clear the infections within 24 hours.
What’s more impressive is E.coli didn’t develop any resistance to this new drug during one-month of treatment. To put this into context, bacteria start developing resistance to the existing antibiotic named ciprofloxacin within a day, and after one month, it becomes 200 times more resistant.
In the coming years, researchers will work with pharmaceutical companies or nonprofit organizations to further examine halicin and develop it for use in humans.