With the help of artificial intelligence, MIT researchers have designed new antibiotics capable of fighting two difficult-to-treat infections: Neisseria gonorrhoeae and multi-resistant Staphylococcus aureus (MRSA).
Using generative AI algorithms, the research team designed more than 36 million possible compounds and computationally analyzed them for their antimicrobial properties. The top candidates they discovered are structurally distinct from all existing antibiotics and appear to act through novel mechanisms that disrupt bacterial cell membranes.
This approach allowed researchers to generate and evaluate theoretical compounds never before seen – a strategy they now hope to apply to identify and design compounds with activity against other species of bacteria.
“We are excited about the new possibilities this project opens up for antibiotic development. Our work shows the power of AI from a drug design perspective and allows us to exploit much larger chemical spaces that were previously inaccessible,” says James Collins, Termeer Professor of Medical Engineering and Sciences in the Institute of Medical Engineering and Sciences (IMES) and the Department of Biological Engineering at MIT, and a member of the Broad Institute.
Collins is the lead author of the study, which appears today In Cell. The paper’s lead authors are MIT postdoc Aarti Krishnan, former postdoc Melis Anahtar ’08, and Jacqueline Valeri PhD ’23.
Exploring chemical space
Over the past 45 years, a few dozen new antibiotics have been approved by the FDA, but most of them are variations of existing antibiotics. At the same time, bacterial resistance to many of these drugs has increased. Globally, drug-resistant bacterial infections are estimated to cause nearly 5 million deaths per year.
Hoping to find new antibiotics to combat this growing problem, Collins and others at MIT Antibiotics-AI Project harnessed the power of AI to examine huge libraries of existing chemical compounds. This work has given rise to several promising drug candidates, including halicin And abaucin.
To build on this progress, Collins and his colleagues decided to expand their research to molecules not found in any chemical library. By using AI to generate hypothetically possible molecules that either don’t exist or haven’t been discovered, they realized it should be possible to explore a much greater diversity of potential drug compounds.
In their new study, the researchers used two different approaches: first, they directed generative AI algorithms to design molecules based on a specific chemical fragment exhibiting antimicrobial activity, and second, they let the algorithms freely generate molecules, without having to include a specific fragment.
For the fragment-based approach, researchers sought to identify molecules capable of killing N. gonorrhoeaea Gram-negative bacteria responsible for gonorrhea. They began by assembling a library of approximately 45 million known chemical fragments, consisting of every possible combination of 11 atoms of carbon, nitrogen, oxygen, fluorine, chlorine and sulfur, as well as fragments from Enamine’s REadily AccessibLe (REAL) space.
Then, they screened the library using machine learning models that Collins’ lab previously trained to predict antibacterial activity against N. gonorrhoeae. This resulted in almost 4 million fragments. They narrowed this pool by removing any fragments that were thought to be cytotoxic to human cells, had chemical liabilities, and were known to be similar to existing antibiotics. That left them with about 1 million applicants.
“We wanted to get rid of anything that might resemble an existing antibiotic, to help solve the antimicrobial resistance crisis in a fundamentally different way. By venturing into underexplored areas of the chemical space, our goal was to discover new mechanisms of action,” says Krishnan.
Through several rounds of additional experiments and computer analyses, the researchers identified a fragment they called F1 that appeared to have promising activity against N. gonorrhoeae. They used this fragment as a basis to generate additional compounds, using two different generative AI algorithms.
One such algorithm, known as chemically reasonable mutations (CReM), works by starting with a particular molecule containing F1 and then generating new molecules by adding, replacing, or removing atoms and chemical groups. The second algorithm, F-VAE (fragment-based variational autoencoder), takes a chemical fragment and transforms it into a complete molecule. It does this by learning patterns about how fragments are commonly modified, based on its pre-training on over a million molecules in the ChEMBL database.
These two algorithms generated approximately 7 million candidates containing F1, which the researchers then computationally examined for activity against N. gonorrhoeae. That review yielded about 1,000 compounds, and the researchers selected 80 to see if they could be produced by chemical synthesis suppliers. Only two of them could be synthesized, and one of them, named NG1, proved very effective in killing N. gonorrhoeae in a laboratory dish and in a mouse model of drug-resistant gonorrhea infection.
Additional experiments revealed that NG1 interacts with a protein called LptA, a novel drug target involved in bacterial outer membrane synthesis. It appears that the drug works by interfering with membrane synthesis, which is fatal to cells.
Constraint-free design
In a second set of studies, researchers explored the potential of using generative AI to freely design molecules, using Gram-positive bacteria, S. aureus as their target.
Once again, the researchers used CReM and VAE to generate molecules, but this time with no constraints other than the general rules for how atoms can join together to form chemically plausible molecules. Together, the models generated more than 29 million compounds. The researchers then applied the same filters as to N. gonorrhoeae candidates, but focusing on S. aureusultimately reducing the pool to around 90 compounds.
They were able to synthesize and test 22 of these molecules, and six of them showed strong antibacterial activity against multidrug-resistant bacteria. S. aureus grown in a laboratory dish. They also found that the best candidate, named DN1, was able to eliminate a methicillin-resistant virus. S. aureus (MRSA) skin infection in a mouse model. These molecules also appear to interfere with bacterial cell membranes, but with broader effects that are not limited to interaction with a specific protein.
Phare Bio, a non-profit organization that is also part of the Antibiotics-AI project, is currently working to further modify NG1 and DN1 to make them suitable for further testing.
“Together with Phare Bio, we are exploring analogues and also working to advance the best candidates to the preclinical level, through medicinal chemistry work,” explains Collins. “We are also excited to apply the platforms developed by Aarti and her team to other bacterial pathogens of interest, including Mycobacterium tuberculosis And Pseudomonas aeruginosa.”
The research was funded in part by the U.S. Defense Threat Reduction Agency, the National Institutes of Health, the Audacious Project, Flu Lab, the Sea Grape Foundation, Rosamund Zander and Hansjorg Wyss for the Wyss Foundation, as well as an anonymous donor.