February 20, 2020 -- To address antibiotic resistance, researchers have developed a machine-learning approach that can search millions of known chemicals to find new potential antimicrobial compounds. This research, published in Cell on February 20, uncovered several promising antibiotic candidates that will move into clinical testing.
After training a deep neural network to identify potential antibiotics that kill bacteria using different mechanisms than current drugs, a team of researchers led by James Collins, PhD, from Massachusetts Institute of Technology (MIT) discovered a new antibiotic compound that killed many challenging pathogens.
"We wanted to develop a platform that would allow us to harness the power of artificial intelligence to usher in a new age of antibiotic drug discovery," Collins, the Termeer Professor of Medical Engineering and Science in MIT's Institute for Medical Engineering and Science and Department of Biological Engineering, said in a statement. "Our approach revealed this amazing molecule, which is arguably one of the more powerful antibiotics that has been discovered."
Need for new methods
Discovering new antibiotics is becoming increasingly difficult. Historically, antibiotics have been discovered by screening soil-dwelling microbes for secondary metabolites that prevented the growth of bacteria. Engineering next-generation versions of existing antibiotics has not been fruitful, and searching synthetic chemical libraries is a costly endeavor that has resulted in no new clinical antibiotics since the implementation of high-throughput screening in the 1980s.
"We're facing a growing crisis around antibiotic resistance, and this situation is being generated by both an increasing number of pathogens becoming resistant to existing antibiotics and an anemic pipeline in the biotech and pharmaceutical industries for new antibiotics," Collins said.
A new pipeline
To help in the fight to find new antibiotics, the MIT researchers developed a novel approach to antibiotic discovery using machine learning to identify novel structural classes of antibiotics. First, they trained a deep neural network model to predict which molecules can block Escherichia coli (E. coli) growth. The training database contained over 1,700 FDA-approved drugs and a set of 800 natural products with diverse structures and a wide range of bioactivities.
Second, they applied the model to chemical libraries (of over 107 million molecules) to identify potential lead compounds. And lastly, selected lead candidates were chosen based on prediction score threshold, chemical structure, and availability.
Chemical database screening led the researchers to identify nine antibacterial compounds that are structurally distant from known antibiotics. By searching the Broad Institute's Drug Repurposing Hub, they identified c-Jun N-terminal kinase inhibitor SU3327, called halicin, as structurally divergent from conventional antibiotics and a potent inhibitor of E. coli growth.
Further testing revealed that halicin provided protection against a wide spectrum of pathogens. Among them were Clostridioides difficile, pan-resistant Acinetobacter baumannii, and Mycobacterium tuberculosis. Of note, the only bacteria the drug was not effective on was Pseudomonas aeruginosa, a difficult-to-treat lung pathogen.
The researchers also tested the effectiveness of halicin in living animals by treating mice with A. baumanni, which has caused infection in American soldiers in Iraq and Afghanistan and is resistant to all known antibiotics. They found that halicin cleared the infection in mice within 25 hours.
While the model was agnostic to mechanism of action, the researchers wanted to elucidate halicin's inhibition mechanisms. RNA sequencing and whole-transcriptome sequencing revealed downregulation of genes involved in cell motility and upregulation of genes required for iron homeostasis. Biochemical assays showed halicin caused disruption of concentration gradients in bacteria. This indicates that disruption of electrochemical transmembrane gradients causes bacterial cell death.
The researchers plan to pursue further studies of halicin in hopes of developing it for use in humans.
After applying the model to the Drug Repurposing Hub, the researchers explored compounds in two additional chemical libraries: the WuXi antituberculosis library and the ZINC15 database. The ZINC15 library contains around 1.5 billion chemical compounds. This screen, which took only three days, identified 23 candidates that were structurally dissimilar from existing antibiotics and predicted to be nontoxic to human cells.
From molecules that showed antibacterial activity during the screen, the researchers found two particularly powerful compounds. They plan to use their model to design new antibiotics and to optimize existing molecules. The model will be an important tool for drug discovery in the future and will help scientists expand antibiotic arsenals and outpace the dissemination of resistance.
Do you have a unique perspective on your research related to microbiology? Contact the editor today to learn more.