Using generative artificial intelligence (generative AI), scientists from Massachusetts Institute of Technology (MIT) and their collaborators elsewhere designed novel antibiotics capable of combating two challenging infections: drug-resistant Neisseria gonorrhoeae and multi-drug-resistant Staphylococcus aureus (MRSA). Details of the work are published in a new Cell paper titled “A generative deep learning approach to de novo antibiotic design.”
According to the paper, the scientists used generative AI to design more than 36 million possible compounds and then computationally screened them for antimicrobial properties. Of that list, the top candidates are structurally distinct from existing antibiotics, and they seem to use novel mechanisms that disrupt bacterial cell membranes.
“We’re excited about the new possibilities that this project opens up for antibiotics development,” said James Collins, PhD, the study’s senior author and a professor of medical engineering and science and professor of biological engineering at MIT. “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.”
This research builds on progress made by Collins and others at MIT’s Antibiotics-AI Project, which, as the name implies, is using AI to search chemical compound libraries. For their study, Collins and his collaborators moved beyond searching these libraries to generating hypothetically possible molecules that either do not exist or have not been discovered.
Specifically, they deployed two different approaches. First, they directed the algorithms to design molecules based on a specific chemical fragment that showed antimicrobial activity. Second, they directed the algorithms to freely generate molecules without requiring them to include a specific fragment.
The team used the fragment-directed approach to identify molecules that could potentially kill N. gonorrhoeae, the bacterium that causes gonorrhea. Their first step was to assemble a library of about 45 million known chemical fragments, consisting of all possible combinations of 11 atoms of carbon, nitrogen, oxygen, fluorine, chlorine, and sulfur, along with fragments from Enamine’s REAL Space collection of compounds.
They then screened the library of molecules using machine learning models trained to predict antibacterial activity against the bacteria of interest. The result? Nearly four million fragments. The researchers narrowed the pool by eliminating fragments predicted to be cytotoxic to human cells, those that displayed chemical liabilities, and those known to be similar to existing antibiotics. That left about a million compounds. Following further experiments and computational analysis, the scientists honed in on a fragment they dubbed F1 that appeared to have promising activity against N. gonorrhoeae.
They used this fragment as the basis for generating additional compounds, using two different generative AI algorithms: CReM, which uses a particular molecule containing F1 as a starting point for generating new molecules by adding, replacing, or deleting atoms and chemical groups, and F-VAE which takes a chemical fragment and builds it into a complete molecule by learning patterns of how fragments are commonly modified.
The two algorithms generated a combined seven million candidates containing F1, which the researchers then computationally screened for activity against N. gonorrhoeae. This narrowed the list down to about 1,000 compounds. From this shortlist, the scientist selected 80 for synthesis. Only two could be synthesized, including one dubbed NG1, which effectively killed N. gonorrhoeae in both cells and a mouse model of drug-resistant gonorrhea.
Looking deeper, the scientists discovered that NG1 interacts with a protein called LptA, a novel drug target involved in the synthesis of the bacterial outer membrane. The drug works by interfering with membrane synthesis, which is fatal to cells.
In a separate group of studies, the scientists used the unconstrained design approach to design novel molecules that target drug-resistant S. aureus. They used the same algorithms as the fragment-based approach, but with no constraints other than general rules that govern how atoms combine to form chemically plausible molecules. In this case, the models generated 29 million compounds.
The team then filtered these down to about 90 compounds. They were able to synthesize and test 22 of these molecules. Six showed strong antibacterial activity against multidrug-resistant S. aureus in cells. Their tests also showed that the top candidate, DN1, successfully treated a MRSA skin infection in a mouse model. Like NG1, DN1 also seems to interfere with bacterial cell membranes, but its interactions are not limited to a specific protein.
Phare Bio, a nonprofit that is also part of the Antibiotics-AI Project, is now working on further modifying NG1 and DN1 to make them suitable for additional testing. Meanwhile, the scientists are aiming to tackle other challenging bacteria, including Mycobacterium tuberculosis and Pseudomonas aeruginosa, Collins said.
The post AI-Designed Molecules Show Potent Activity Against Resistant Bacteria appeared first on GEN – Genetic Engineering and Biotechnology News.
