They concluded that it should be feasible to investigate a far wider variety of prospective medication compounds by utilizing AI to create hypothetically possible molecules that are either nonexistent or have not yet been found.
The researchers used two distinct strategies in their new study: first, they instructed generative AI algorithms to create molecules based on a particular chemical fragment that exhibited antimicrobial activity, and second, they allowed the algorithms to create molecules independently without requiring the inclusion of a particular fragment.
The goal of the fragment-based strategy was to find compounds that would kill the Gram-negative bacterium that causes gonorrhea, N. gonorrhoeae. They started by compiling a library of over 45 million known chemical fragments, including fragments from Enamine’s Readily Accessible (REAL) space and every possible combination of the 11 atoms of carbon, nitrogen, oxygen, fluorine, chlorine, and sulfur.
A novel medication that targets the bacterium that causes gonorrhea, N. gonorrhoeae, has been produced by researchers. After testing a library of
antibiotic fragments using machine-learning methods, they discovered a fragment known as F1 that showed encouraging activity. To create chemicals, they used two AI algorithms: F-VAE (fragment-based variational autoencoder) and chemically reasonable mutations (CReM).
After computationally screening over 7 million candidates, the team produced roughly 1,000 molecules. N. gonorrhoeae was effectively killed by NG1, one of the two that could be produced. A protein known as LptA, a novel therapeutic target involved in the production of the bacterial outer membrane, interacts with NG1.
In a second round, Researchers have used generative AI to design molecules targeting Gram-positive bacteria, S. aureus. They generated over 29 million compounds using CReM and VAE, narrowing the pool down to 90. 22 of these molecules were synthesized and tested, with six showing strong antibacterial activity against multi-drug-resistant S. aureus. The top candidate, DN1, cleared a methicillin-resistant S. aureus skin infection in a mouse model. Phare Bio, a nonprofit part of the Antibiotics-AI Project, is working on further modifying NG1 and DN1 for further testing.
