Researchers in North America have made a groundbreaking discovery in antibiotic development by utilizing machine learning. The team has identified a new compound called abaucin, which has shown promising results in controlling bacterial infections.
Initially developed as an anti-diabetes drug, the compound demonstrated the ability to inhibit the growth of Acinetobacter baumannii, a multidrug-resistant bacterium commonly found in hospitals and known to cause severe conditions such as pneumonia and meningitis.
The study published in Nature Chemical Biology and as reported by MIT News, led by Jonathan Stokes, an assistant professor of biochemistry and biomedical sciences at McMaster University in Canada, employed machine learning techniques to screen approximately 7,500 molecules.
This extensive screening process identified abaucin as a potential candidate for combating Acinetobacter baumannii. Further investigations revealed that abaucin affects the trafficking of lipoproteins through a mechanism involving a protein called LolE, which contributes to the movement of lipoproteins from the inner to the outer membrane.
In addition to inhibiting bacterial growth in vitro, abaucin demonstrated the ability to control Acinetobacter baumannii infections in a mouse wound model. The researchers emphasized the significance of their findings, highlighting the role of machine learning in discovering new antibiotics.
The recent breakthrough builds upon the team’s previous research, which showcased the potential of machine learning in identifying new antibacterial molecules targeting E. coli. For the current study, the researchers utilized a message-passing deep neural network to identify novel antibiotics effective against Acinetobacter baumannii.
To train their computational model, the team exposed the bacterium to over 7,600 compounds, including off-patent drugs and synthetic chemicals. Through this process, they identified 480 “active” molecules that affected bacterial growth, while 7,204 were labeled “inactive.” The dataset was then used to train a binary classifier capable of predicting the activity of structurally new molecules against Acinetobacter baumannii.
Following the training phase, the researchers employed the model to analyze over 6,600 compounds that had yet to be encountered previously. These compounds were sourced from the Drug Repurposing Hub at the Broad Institute. The analysis, which took less than two hours, yielded several hundred potential hits, of which 240 were experimentally tested in the laboratory.
The researchers specifically focused on compounds that differed from existing antibiotics or molecules in the training data. Ultimately, nine antibiotics were discovered, with abaucin proving highly potent against Acinetobacter baumannii while not affecting other bacterial species such as Pseudomonas aeruginosa, Staphylococcus aureus, and carbapenem-resistant Enterobacteriaceae.
The narrow-spectrum action of abaucin is particularly advantageous as it reduces the likelihood of bacteria developing resistance to the antibiotic. Furthermore, it avoids disrupting the balance of beneficial bacteria during treatment, as it explicitly targets Acinetobacter baumannii. Additional investigations revealed that abaucin effectively treated mouse wound infections caused by the bacterium, including various drug-resistant strains isolated from human patients.
Despite the significant progress made, the researchers are still determining why abaucin displays such selectivity for Acinetobacter baumannii, considering that all Gram-negative bacteria express the LolE enzyme. Stokes speculates that the unique lipoprotein trafficking mechanism employed by Acinetobacter baumannii may be responsible for this narrow spectrum of activity. The team continues their experimental data acquisition to gain further insights into this phenomenon.
The discovery of abaucin through machine learning represents a significant step forward in antibiotic development. By harnessing the power of artificial intelligence, researchers have unlocked new possibilities for identifying potent antibiotics with specific activity against challenging bacterial pathogens. This innovative approach can potentially revolutionize the discovery process and contribute to the ongoing fight against antibiotic resistance.
