Researchers at MIT have harnessed the power of generative artificial intelligence to make a major leap in our battle against antibiotic resistance. The team has devised a novel approach using AI to develop unique antibiotics that show promising results against two notoriously drug-resistant bacterial infections: Neisseria gonorrhoeae and the multi-drug-resistant Staphylococcus aureus (MRSA). What sets this project apart and adds to its brilliance is the fact that the new antibiotic compounds are structurally unique, working through never-before-seen mechanisms.
Thanks to AI, the MIT team was able to generate over 36 million potential chemical compounds and used an advanced screening process to examine each for antimicrobial activity. James Collins, the mind powering this initiative, stated with palpable excitement, “Our work shows the power of AI from a drug design standpoint, and enables us to exploit much larger chemical spaces that were previously inaccessible.”
Historically, methods for discovering new antibiotics largely produced slightly modified variants of existing drugs. Given bacteria’s ability to develop resistance against these modified drugs, this posed a massive problem causing an estimated 5 million fatalities annually. However, by leveraging AI to screen chemical databases, the MIT Antibiotics-AI Project has produced promising candidates such as halicin and abaucin.
Deviating further from the norm, in their most recent study, Collins’ team ventured into uncharted chemical space. Using AI, they successfully invented entirely new molecules, which do not exist in any current databases or libraries. The focus of their study was twofold: first, targeting the bacterium responsible for gonorrhea, N. gonorrhoeae; and second, combating S. aureus, a bacterium infamous for resisting multiple antibiotics.
To confront N. gonorrhoeae, the team commenced with a library of 45 million known chemical fragments, enlisting machine learning models to filter them down to just 1 million candidates. The objective was to exclude any compounds resembling existing antibiotics or any with predicted toxicity to human cells. One compound, dubbed ‘NG1’, resulted from this process and demonstrated powerful activity against N. gonorrhoeae in laboratory settings and a mouse model.
The approach adopted for S. aureus differed, with the AI given total freedom to design molecules without any predetermined fragments. Here, the standout candidate, ‘DN1’, indicated powerful antibacterial activity and was successful in treating MRSA skin infection in mice.
As Collins and his team continue their pioneering research, Phare Bio, a partner of the Antibiotics AI Project, has set out to refine both NG1 and DN1 for further tests. At the same time, the MIT team plans to turn the spotlight of their AI platform on a variety of dangerous pathogens, including Mycobacterium tuberculosis and Pseudomonas aeruginosa.
This revolutionary work is supported by a range of organizations, including the U.S. Defense Threat Reduction Agency, the National Institutes of Health, the Audacious Project, and several private donors. You can delve deeper into the original study at MIT News.
This website uses cookies.