Scientists are using machine learning techniques to sift through millions of molecules in search of effective new antibiotics. In a new paper published on 20 February in the journal Cell, the authors describe a new molecule, which is the first-ever antibiotic to be discovered by artificial intelligence (1).
The team of researchers, led by Prof Jim Collins of the Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering at the Massachusetts Institute of Technology (MIT), used advanced machine-learning methods to pinpoint new types of antibiotics among millions of molecules and found that one that is effective against a range of bacteria, including a previously untreatable strain of tuberculosis.
The algorithm is based on something called a neural network — inspired by the neural network of the human brain ― widely used in machine learning. Here, the method was used to identify molecules that prevent the growth of Escherichia coli. First, the researchers trained the model using more than 2,335 molecules with known antibiotic activity, including both approved antibiotics and natural products with diverse structures and a wide range of bioactivities.
Then the model was used to screen more than 6000 candidate molecules in the Drug Repurposing Hub — an open-access collection of compounds, many of which are FDA-approved, created by researchers at the Broad Institute in the hopes of finding new uses for existing drugs.
The model discovered one molecule, in particular, with potentially strong antibacterial activity, but a chemical structure different from existing antibiotics. The molecule, named halicin — after HAL, the fictional intelligent computer in the film 2001: A Space Odyssey — was previously investigated as a diabetes treatment. Now, the drug might soon prove to be a powerful antibiotic.
In the laboratory, halicin killed several strains of highly problematic disease-causing bacteria, including Clostridium difficile, Acinetobacter baumannii, and Mycobacterium tuberculosis. Then, when the scientists tested the drug against a strain of A. baumannii that is resistant to all known antibiotics, infections in mice were completely cleared within 24 hours.
Preliminary work showed the antibiotic works by disrupting the bacteria’s ability to maintain an electrochemical gradient across their cell membranes necessary for many important functions, like storing energy. And this mode of death could be difficult for bacterial cells to develop resistance to, according to the authors. In their study, E. coli did not develop any resistance to halicin during a 30-day treatment period.
“When you’re dealing with a molecule that likely associates with membrane components, a cell can’t necessarily acquire a single mutation or a couple of mutations to change the chemistry of the outer membrane. Mutations like that tend to be far more complex to acquire evolutionarily,” said lead author Dr Jonathan Stokes.
The researchers are now hoping to team up with a pharmaceutical company or non-profit to test the new antibiotic in clinical trials.
They also used the model to screen more than 100 million molecules from the ZINC15 database — an online collection of around 1.5 billion chemical compounds — and identified 23 candidates that are structurally dissimilar from existing antibiotics. Eight of the molecules showed antibacterial activity, and two were quite potent — which they plan to test further.
With the rise of bacterial resistance, existing antibiotics are becoming less effective. Antibiotic-resistant infections could kill up to 10 million people per year by 2050, according to a recent report released in April 2019 by the United Nations Interagency Coordination Group (IACG) on Antimicrobial Resistance.
“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 anaemic pipeline in the biotech and pharmaceutical industries for new antibiotics”, said Collins, in a statement. Moreover, not many new antibiotics have been developed over the past few decades, and most of the newly approved antibiotics are simply a variation of existing drugs.
The machine learning approach used by Collins and his colleagues differs from previous approaches because instead of looking for specific structures or classes of molecules, the artificial intelligence is trained to identify molecules with a certain activity, in this case, antibiotic activity.
“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 said in a statement. He added: “Our approach revealed this amazing molecule which is arguably one of the more powerful antibiotics that has been discovered.”
(1) Stokes, J. M. et al. A Deep Learning Approach to Antibiotic Discovery. Cell (2020). DOI: 10.1016/j.cell.2020.01.021
Image: Strain of Mycobacterium tuberculosis that causes untreatable tuberculosis infection.