The study showcases the potential of AI in antibiotic discovery, offering insights into antibiotic design and aiding in the development of novel drug candidates.
MIT researchers have harnessed the power of artificial intelligence (AI) to discover a new class of compounds with the potential to combat methicillin-resistant Staphylococcus aureus (MRSA), a drug-resistant bacterium causing over 10,000 deaths annually in the United States. The breakthrough, detailed in a Nature study, demonstrates the compounds’ effectiveness against MRSA in lab experiments and two mouse models, with minimal toxicity to human cells, making them promising candidates for drug development.
Understanding AI’s Predictions for Antibiotic Potency
A key aspect of the study is the transparency in understanding how the deep-learning model makes antibiotic potency predictions. The researchers unveiled the information utilized by the model to enhance antibiotic design. The team fed the model information on the compounds’ antibiotic properties and chemical structures by training a deep-learning model with an expanded dataset of around 39,000 compounds tested for antibiotic activity against MRSA.
Employing an algorithm called Monte Carlo tree search, typically used for enhancing explainability in deep learning models, the researchers estimated each molecule’s antimicrobial activity and predicted the substructures contributing to that activity. This knowledge aids in pinpointing features that make certain molecules effective antibiotics, facilitating the design of improved drugs.
See also: Researchers Use AlphaFold To Design Cancer Drug
Discovery and Testing of Promising Antibiotic Candidates
Through a comprehensive screening process involving three additional deep-learning models to assess toxicity to human cells, the researchers identified compounds capable of killing microbes while posing minimal harm to human cells. From a pool of 12 million commercially available compounds, the models pinpointed five different classes predicted to be active against MRSA based on chemical substructures.
Further experiments narrowed down the selection to two promising antibiotic candidates from the same class. Testing against MRSA in lab dishes and mouse models demonstrated significant reductions in MRSA populations. The compounds disrupt bacterial cell membranes, selectively attacking Gram-positive pathogens like MRSA without substantial damage to human cell membranes.
The researchers shared their findings with Phare Bio, a nonprofit associated with the Antibiotics-AI Project, for further analysis. The project aims to discover new antibiotic classes against deadly bacteria over seven years. The study showcases the potential of AI in antibiotic discovery, offering insights into antibiotic design and aiding in the development of novel drug candidates.
