Novel Antibiotics Powered by AI: A Breakthrough at MIT
With the relentless rise of antibiotic-resistant infections, innovative solutions are urgently needed. Researchers at MIT have harnessed the power of artificial intelligence to design novel antibiotics aimed at two particularly troublesome adversaries: drug-resistant Neisseria gonorrhoeae and multi-drug-resistant Staphylococcus aureus (MRSA).
The Role of Generative AI in Antibiotic Discovery
In a pioneering effort, the MIT research team utilized generative AI algorithms to create over 36 million chemical compounds, each screened for their potential antimicrobial properties. The top candidates identified through this approach are not only structurally distinct from existing antibiotics but also operate through novel mechanisms, effectively disrupting bacterial cell membranes. This groundbreaking work underscores the transformative potential of AI in drug discovery, allowing researchers to explore chemical spaces previously thought inaccessible.
James Collins, senior author and Termeer Professor of Medical Engineering and Science at MIT, expressed enthusiasm about the project’s implications. “We’re excited about the new possibilities that this project opens up for antibiotics development,” he shared. The research encompasses collaboration among postdoctoral researchers Aarti Krishnan, Melis Anahtar, and Jacqueline Valeri, and has been published in Cell.
The Challenge of Antibiotic Resistance
Antibiotic resistance is growing at an alarming rate, with nearly 5 million deaths attributed to drug-resistant bacterial infections each year. Although the FDA has approved a limited number of new antibiotics over the past 45 years, most are merely variants of existing medications. Collins and his team at the MIT Antibiotics-AI Project aim to combat this crisis through innovative strategies, moving beyond established compounds to explore entirely new molecular candidates.
Exploring Chemical Space
To discover new antibiotics, researchers expanded their focus to explore molecules that do not exist in current chemical libraries. This approach involved two main strategies: fragment-based design and free molecule generation.
In the fragment-based design, the team began with a library of approximately 45 million known chemical fragments. By applying machine-learning models trained on previous antibacterial activity against N. gonorrhoeae, they filtered the candidates, ultimately narrowing the list down to about 1 million fragments that were non-cytotoxic and chemically unique.
Selecting Effective Compounds
Through rigorous computational analysis, the researchers identified a promising fragment known as F1. They utilized two generative AI algorithms to develop new compounds based on this fragment. The first algorithm, chemically reasonable mutations (CReM), modified an initial molecule containing F1, while the second algorithm, F-VAE (fragment-based variational autoencoder), constructed a complete molecule from the fragment.
Together, these algorithms yielded roughly 7 million potential candidates. After further screening, around 1,000 compounds were identified for synthesis, with only two proving feasible. One of these, named NG1, demonstrated significant efficacy against N. gonorrhoeae in lab settings and in mouse models.
Target and Mechanism of Action
NG1 interacts with a specific protein called LptA, crucial for bacterial outer membrane synthesis. The interference with membrane synthesis caused by NG1 is lethal to the bacteria, offering a novel mechanism of action.
Unconstrained Design for Broader Applications
For the second round of research, the team turned their attention to S. aureus, utilizing a completely unconstrained generative AI design process. By applying CReM and VAE algorithms without specific fragment requirements, they generated an impressive 29 million compounds. After applying similar filtering criteria as before, they narrowed the candidates down to about 90.
From this selective pool, they synthesized and tested 22 molecules, with six exhibiting strong antibacterial properties against MRSA. The standout candidate, DN1, was successful in clearing MRSA skin infections in mouse models and appears to compromise bacterial membranes through broader interactions beyond a single protein target.
Collaboration and Future Directions
The collaboration with Phare Bio, a nonprofit involved in the Antibiotics-AI Project, is crucial for advancing both NG1 and DN1. The two compounds will undergo medicinal chemistry modifications to enhance their effectiveness and readiness for broader testing.
Collins shares their aspirations, stating that they are excited to apply the developed platforms to target other significant bacterial pathogens, including Mycobacterium tuberculosis and Pseudomonas aeruginosa.
Funding and Support
The groundbreaking research at MIT has garnered significant support from various organizations, including the U.S. Defense Threat Reduction Agency, the National Institutes of Health, and the Audacious Project, among others. This collaborative effort highlights the commitment to tackling antibiotic resistance through innovative research and technology.