Researchers at MIT McMaster University have identified a new antibiotic that can dispatch a type of bacteria responsible for many drug-resistant infections using an algorithm based on machine learning.
If the drug is formulated for patient usage, it could potentially aid in combating Acinetobacter baumannii, a bacterial species commonly present in healthcare facilities and associated with severe infections such as pneumonia, meningitis, and other critical ailments. Furthermore, this microorganism is a primary cause of infections among injured military personnel in Iraq and Afghanistan back then.
Jonathan Stokes, a former MIT postdoc who is now an Assistant Professor of Biochemistry and Biomedical Sciences at McMaster University, explained, “Acinetobacter has the ability to persist on hospital doorknobs and equipment for extended durations, and it can acquire antibiotic resistance genes from its surroundings. It is increasingly prevalent to encounter A. baumannii strains that resist nearly all available antibiotics.”
The scientists employed a machine-learning model to examine approximately 7,000 potential drug compounds and identify the new medication. The model was trained to assess whether a chemical compound could effectively hinder the growth of A. baumannii.
James Collins, the Termeer Professor of Medical Engineering and Science at MIT’s Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering, said, “This discovery further reinforces the notion that AI has the potential to greatly accelerate and expand our quest for innovative antibiotics. I am thrilled that this research demonstrates the utility of AI in addressing challenging pathogens like A. baumannii.”
During the initial experiment, the researchers educated a machine-learning algorithm to recognise chemical compositions with the potential to impede the growth of E. coli. Employing this algorithm, they screened over 100 million compounds and discovered a molecule called halicin. Their investigations demonstrated that halicin could eliminate not only E. coli but various other bacterial strains that resist conventional treatment methods.
To gather training data for their computational model, the researchers initially exposed A. baumannii cultivated in a laboratory dish to approximately 7,500 distinct chemical compounds to identify those capable of impeding the microbe’s growth. Subsequently, they input the structure of each molecule into the model while providing information regarding whether each structure could inhibit bacterial growth. It enabled the algorithm to grasp the chemical characteristics associated with growth inhibition.
Once the model was trained, the researchers employed it to analyse a set of 6,680 compounds, which had not been previously encountered by the model. This analysis selected several hundred top candidate compounds within less than two hours. Among these candidates, the researchers handpicked 240 for experimental laboratory testing, focusing on compounds possessing distinct structures from existing antibiotics or molecules within the training data.
From the test’s results, the researchers discovered nine antibiotics, one of which exhibited exceptional potency. Interestingly, this particular compound, initially investigated as a potential treatment for diabetes, showcased remarkable efficacy in eliminating A. baumannii while having no impact on other bacterial species such as Pseudomonas aeruginosa, Staphylococcus aureus, and carbapenem-resistant Enterobacteriaceae.
The ability of this antibiotic to exhibit a narrow-spectrum killing effect is a highly desirable characteristic as it reduces the risk of bacteria swiftly developing resistance to the drug. Additionally, this attribute offers the advantage of potentially sparing beneficial bacteria in the human gut, which play a crucial role in suppressing opportunistic infections like Clostridium difficile.