In a groundbreaking discovery, MIT scientists have leveraged advanced machine learning algorithms to identify a potent new antibiotic compound. During extensive laboratory evaluations, this revolutionary drug demonstrated remarkable efficacy against numerous problematic disease-causing bacteria worldwide, including several strains that exhibit resistance to all currently available antibiotics. Furthermore, it successfully eliminated infections in two distinct mouse models, showcasing its potential as a game-changing medical treatment.
The sophisticated computer model employed in this research can rapidly analyze more than 100 million chemical compounds within just a few days. This innovative technology is specifically engineered to identify potential antibiotics that combat bacteria through mechanisms distinct from those utilized by existing medications.
"Our objective was to create a platform that would enable us to harness the transformative power of artificial intelligence to revolutionize antibiotic drug discovery," explains James Collins, the Termeer Professor of Medical Engineering and Science in MIT's Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering. "This approach led us to uncover this extraordinary molecule, which stands as one of the most significant antibiotic breakthroughs in recent history."
Throughout their investigation, the research team also pinpointed several additional promising antibiotic candidates that they intend to examine more thoroughly in subsequent studies. They are confident that their model could also be instrumental in designing novel medications, leveraging its acquired knowledge about chemical structures that enable compounds to effectively eliminate harmful bacteria.
"The machine learning model can explore, in silico, vast chemical spaces that would be prohibitively expensive for traditional experimental approaches to analyze," notes Regina Barzilay, the Delta Electronics Professor of Electrical Engineering and Computer Science in MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL).
Barzilay and Collins, who serve as faculty co-leads for MIT's Abdul Latif Jameel Clinic for Machine Learning in Health (J-Clinic), are the senior authors of the study, which was published today in Cell. The paper's lead author is Jonathan Stokes, a postdoctoral researcher at MIT and the Broad Institute of MIT and Harvard.
A Revolutionary Approach
Over the last several decades, the development of new antibiotics has been remarkably limited, with most recently approved medications representing only minor variations of existing drugs. Current methodologies for screening potential antibiotics are often exorbitantly expensive, demand substantial time investments, and typically explore only a limited spectrum of chemical diversity.
"We're confronting an escalating crisis regarding antibiotic resistance, exacerbated both by the increasing number of pathogens developing resistance to existing treatments and by the inadequate pipeline for new antibiotics within the biotech and pharmaceutical sectors," Collins emphasizes.
To discover entirely novel compounds, Collins collaborated with Barzilay, Professor Tommi Jaakkola, and their students Kevin Yang, Kyle Swanson, and Wengong Jin, who had previously created machine learning models capable of analyzing the molecular structures of compounds and correlating them with specific characteristics, such as antibacterial properties.
The concept of employing predictive computer models for "in silico" screening is not entirely new; however, previous models lacked the precision required to revolutionize drug discovery. Earlier approaches represented molecules as vectors indicating the presence or absence of certain chemical groups. In contrast, the new neural networks can automatically learn these representations, mapping molecules into continuous vectors that are subsequently used to predict their properties.
For this research, the scientists designed their model to identify chemical features that enable molecules to effectively combat E. coli. To accomplish this, they trained the model using approximately 2,500 molecules, including about 1,700 FDA-approved drugs and a collection of 800 natural products with diverse structures and a broad spectrum of bioactivities.
After completing the training process, the researchers tested their model on the Broad Institute's Drug Repurposing Hub, which contains a library of about 6,000 compounds. The model identified one molecule predicted to exhibit potent antibacterial activity and possessing a chemical structure distinct from any existing antibiotics. Using a separate machine learning model, the researchers also determined that this molecule would likely have minimal toxicity to human cells.
This molecule, which the researchers named halicin in homage to the fictional artificial intelligence system from "2001: A Space Odyssey," had previously been investigated as a potential diabetes treatment. The researchers tested it against dozens of bacterial strains isolated from patients and cultivated in laboratory settings, discovering that it effectively eliminated many treatment-resistant strains, including Clostridium difficile, Acinetobacter baumannii, and Mycobacterium tuberculosis. The drug proved effective against nearly every species tested, with the exception of Pseudomonas aeruginosa, a notoriously difficult-to-treat lung pathogen.
To evaluate halicin's effectiveness in living organisms, the researchers administered it to mice infected with A. baumannii, a bacterium that has affected numerous U.S. soldiers stationed in Iraq and Afghanistan. The particular strain of A. baumannii used in the experiment was resistant to all known antibiotics, yet application of a halicin-containing ointment completely eradicated the infections within 24 hours.
Initial investigations suggest that halicin eliminates bacteria by disrupting their capacity to maintain an electrochemical gradient across their cell membranes. This gradient serves multiple critical functions, including producing ATP (molecules that cells use to store energy). When this gradient breaks down, the cells perish. According to the researchers, this mechanism of action could present significant challenges for bacteria attempting to develop resistance.
"When dealing with a molecule that likely interacts with membrane components, a cell can't necessarily acquire a single mutation or a couple of mutations to alter the chemistry of the outer membrane," Stokes explains. "Mutations of this nature tend to be considerably more complex to acquire evolutionarily."
In their study, the researchers observed that E. coli did not develop any resistance to halicin throughout a 30-day treatment period. In stark contrast, the bacteria began developing resistance to the antibiotic ciprofloxacin within just one to three days. After 30 days, the bacteria were approximately 200 times more resistant to ciprofloxacin than at the beginning of the experiment.
The research team intends to conduct additional studies of halicin, potentially collaborating with a pharmaceutical company or nonprofit organization, with the goal of developing it for human therapeutic applications.
Optimized Molecular Structures
Following the identification of halicin, the researchers also utilized their model to screen more than 100 million molecules selected from the ZINC15 database, an online repository containing approximately 1.5 billion chemical compounds. This screening process, which required only three days, identified 23 candidates with structures dissimilar from existing antibiotics and predicted to have low toxicity to human cells.
In laboratory tests against five bacterial species, the researchers discovered that eight of the molecules exhibited antibacterial activity, with two showing particularly potent effects. The team now plans to conduct further testing of these molecules and to continue screening additional compounds from the ZINC15 database.
The researchers also intend to employ their model to design new antibiotics and optimize existing molecules. For instance, they could train the model to incorporate features that would enable a particular antibiotic to target only specific bacteria, thereby preventing it from eliminating beneficial bacteria in a patient's digestive system.
"This pioneering work represents a paradigm shift in antibiotic discovery and indeed in drug discovery more broadly," states Roy Kishony, a professor of biology and computer science at Technion (the Israel Institute of Technology), who was not involved in the research. "Beyond in silico screens, this approach will facilitate the application of deep learning throughout all stages of antibiotic development, from initial discovery to enhanced efficacy and reduced toxicity through drug modifications and medicinal chemistry."
The research received funding from the Abdul Latif Jameel Clinic for Machine Learning in Health, the Defense Threat Reduction Agency, the Broad Institute, the DARPA Make-It Program, the Canadian Institutes of Health Research, the Canadian Foundation for Innovation, the Canada Research Chairs Program, the Banting Fellowships Program, the Human Frontier Science Program, the Pershing Square Foundation, the Swiss National Science Foundation, a National Institutes of Health Early Investigator Award, the National Science Foundation Graduate Research Fellowship Program, and a gift from Anita and Josh Bekenstein.
