Artificial intelligence (AI) has been used to screen almost 7000 chemical compounds, identifying a potent and novel class of antibiotics capable of killing one of the most problematic superbugs.
The researchers focused on Acinetobacter baumannii, a hospital-acquired bacterial pathogen that causes meningitis, pneumonia and blood infections and which is often resistant to many current antibiotics. It is one of the top three superbugs the WHO deems a ‘critical threat’.
To find the new antibiotic, the team from Canada’s McMaster University and the Massachusetts Institute of Technology, US, (MIT) first had to train the AI systems to identify the chemical features that could attack A. baumannii. They did this in the lab by testing the bacteria’s survival against thousands of drugs with known chemical structures.
This information was then fed into the system and used to analyse and assess the superbug-killing properties of 6680 molecules with unknown effectiveness that the AI had not previously seen.
Within an hour and a half, the AI had produced a shortlist, of which 240 compounds were tested in the lab. Nine potential antibiotics were discovered, with one in particular, abaucin, being particularly potent (Nature Chemical Biology, doi: 10.1038/s41589-023-01349-8).
Abaucin not only treated infected wounds in mice and killed A. baumannii samples from patients – it also displayed activity against A. baumannii isolates resistant to all current clinical antibiotics, says Jonathan Stokes, Assistant Professor of biochemistry and biomedical sciences at McMaster University. The drug was also only specific for A. baumannii and not other species of bacteria.
‘Based on our observations, abaucin inhibits a biological process called lipoprotein trafficking in A. baumannii [the movement of lipoproteins to the outer membrane]. This is a biological process that is not targeted by existing clinical antibiotics,’ Stokes said. At the molecular level, it appears that A. baumannii performs lipoprotein trafficking slightly differently than other bacterial species hence why the drug is specific for this species, he added.
The researchers are now working to optimise the structure of abaucin to enhance its potency and medicinal properties.
Discovering new antibiotics using AI could be a game changer in the race against antimicrobial resistance (AMR). According to the WHO, it took just two-to-three years for bacteria to develop resistance to antibiotics launched between 1970 and 2000. Between 2017 and 2021, just 12 new antibiotics entered the market, while only 27 are in clinical trials and only four of these have new mechanisms of action, the WHO said.
AI could be important in addressing AMR by providing an innovation injection into finding new antibiotics and cutting timelines and R&D costs. In the case of abaucin, the time savings were significant, Stokes said.
‘Our trained model was able to perform predictions on the chemical collection of almost 7000 molecules in roughly an hour and a half. If we screened this collection in the laboratory, it would have taken us about three weeks, maybe a month.’
The more molecules screened the more significant the time saving, he said. ‘We routinely perform AI predictions on the scale of about 100m molecules within about a week. It is not feasible to perform conventional laboratory screens against 100m molecule libraries – if it were possible, it would take years to complete.
‘I believe our study showcases the utility of AI and machine learning (ML) in discovering structurally and functionally novel antibiotics against a truly challenging bacterium.’
Abaucin is not the first antibiotic to be discovered using AI. In 2020, MIT researchers, including Stokes, discovered halicin, which had previously been investigated as a diabetes drug but was found to kill many drug-resistant bacterial species.
A spokesperson from the US pharmaceutical trade body PhRMA told C&I there was ‘great potential’ for using AI and ML in antibiotic drug development. They welcomed efforts from regulators to advance the use of digital health technologies such as AI and for helping to inform the regulatory landscape on the use of the technology in drug development.