“`html
With assistance from artificial intelligence, MIT investigators have created innovative antibiotics capable of tackling two difficult-to-treat infections: drug-resistant Neisseria gonorrhoeae and multidrug-resistant Staphylococcus aureus (MRSA).
Leveraging generative AI algorithms, the research team formulated over 36 million potential compounds and computationally assessed them for antimicrobial properties. The top candidates identified are structurally unique compared to any current antibiotics, and they seem to function through new mechanisms that disrupt bacterial cell membranes.
This strategy permitted the researchers to generate and assess theoretical compounds that have never been encountered before — a method they now aspire to utilize for discovering and creating compounds effective against other bacterial species.
“We’re thrilled about the fresh opportunities this project unveils for antibiotic development. Our research illustrates the strength of AI in drug design and allows us to explore much broader chemical realms that were previously out of reach,” states James Collins, the Termeer Professor of Medical Engineering and Science at MIT’s Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering.
Collins is the lead author of the study, which is published today in Cell. The paper’s primary authors include MIT postdoc Aarti Krishnan, former postdoc Melis Anahtar ’08, and Jacqueline Valeri PhD ’23.
Investigating chemical space
In the last 45 years, only a few dozen new antibiotics have received FDA approval, but most are variations of pre-existing antibiotics. Simultaneously, bacterial resistance to many of these medications has been on the rise. Worldwide, drug-resistant bacterial infections are estimated to cause nearly 5 million deaths annually.
In the quest for new antibiotics to address this escalating issue, Collins and others at MIT’s Antibiotics-AI Project have harnessed AI to screen extensive libraries of existing chemical compounds. This effort has produced several promising drug candidates, including halicin and abaucin.
To build on that advancement, Collins and his team decided to widen their search to molecules not found in any chemical libraries. By employing AI to create hypothetically possible molecules that are non-existent or undiscovered, they recognized the potential to explore a significantly greater diversity of possible drug compounds.
In their latest study, the researchers implemented two distinct approaches: Firstly, they directed generative AI algorithms to design molecules based on a specific chemical fragment exhibiting antimicrobial properties, and secondly, they allowed the algorithms to freely generate molecules without the necessity of including a specific fragment.
For the fragment-focused approach, the researchers aimed to identify molecules capable of eliminating N. gonorrhoeae, a Gram-negative bacterium responsible for gonorrhea. They commenced by compiling a library of approximately 45 million known chemical fragments composed of all conceivable combinations of 11 atoms of carbon, nitrogen, oxygen, fluorine, chlorine, and sulfur, alongside fragments from Enamine’s REadily AccessibLe (REAL) space.
Following this, they screened the library utilizing machine-learning models that Collins’ lab had previously trained to forecast antibacterial activity against N. gonorrhoeae. This yielded nearly 4 million fragments. They refined that pool by eliminating any fragments predicted to be harmful to human cells, exhibited chemical vulnerabilities, or were recognized as resembling existing antibiotics. This left them with roughly 1 million candidates.
“We aimed to eliminate anything that bore resemblance to an existing antibiotic, to tackle the antimicrobial resistance crisis in a fundamentally different manner. By venturing into underexplored regions of chemical space, our objective was to uncover innovative mechanisms of action,” states Krishnan.
Through multiple rounds of further experiments and computational analysis, the researchers identified a fragment named F1 that demonstrated promising activity against N. gonorrhoeae. They utilized this fragment as the foundation for generating additional compounds, employing two different generative AI algorithms.
One of those algorithms, known as chemically reasonable mutations (CReM), operates by starting from a specific molecule containing F1 and subsequently generating new molecules by adding, substituting, or removing atoms and chemical groups. The second algorithm, F-VAE (fragment-based variational autoencoder), takes a chemical fragment and constructs it into a complete molecule. It accomplishes this by learning patterns of how fragments are usually modified, based on its training on over 1 million molecules from the ChEMBL database.
These two algorithms produced approximately 7 million candidates containing F1, which the researchers then computationally screened for activity against N. gonorrhoeae. This screening resulted in about 1,000 compounds, and the researchers chose 80 of those to determine if they could be synthesized by chemical synthesis vendors. Only two of these could be produced, and one of them, designated NG1, proved highly effective at eliminating N. gonorrhoeae in a laboratory setting and in a mouse model of drug-resistant gonorrhea infection.
Further experiments revealed that NG1 interacts with a protein named LptA, a new drug target involved in the creation of the bacterial outer membrane. It seems that the drug disrupts membrane synthesis, leading to cell death.
Unrestricted design
In a second series of studies, the researchers investigated the capacity of using generative AI to freely create molecules targeting Gram-positive bacteria, S. aureus.
Once more, the researchers employed CReM and VAE to generate molecules, but this time without restrictions other than the fundamental rules regarding how atoms can bond to form chemically plausible molecules. Collectively, the models generated over 29 million compounds. The researchers then implemented the same filters applied to the N. gonorrhoeae candidates, but focusing on S. aureus, ultimately narrowing the results down to approximately 90 compounds.
They were able to synthesize and evaluate 22 of these molecules, with six demonstrating robust antibacterial activity against multidrug-resistant S. aureus cultured in a laboratory dish. They also discovered that the top candidate, called DN1, could successfully resolve a methicillin-resistant S. aureus (MRSA) skin infection in a mouse model. These molecules also appear to disrupt bacterial cell membranes, but with a wider range of effects not restricted to interaction with a single protein.
Phare Bio, a nonprofit also involved in the Antibiotics-AI Project, is now working to further modify NG1 and DN1 to prepare them for more in-depth testing.
“In collaboration with Phare Bio, we are investigating analogs, and advancing the most promising candidates preclinically through medicinal chemistry efforts,” shares Collins. “We are also enthusiastic about applying the platforms that Aarti and the team have developed to other bacterial pathogens of concern, particularly Mycobacterium tuberculosis and Pseudomonas aeruginosa.”
The research was partially funded by the U.S. Defense Threat Reduction Agency, the National Institutes of Health, the Audacious Project, Flu Lab, the Sea Grape Foundation, and donors including Rosamund Zander, Hansjorg Wyss for the Wyss Foundation, and an anonymous benefactor.
“`