Researchers from MIT have developed artificial intelligence (AI)-enabled compounds that effectively combat Methicillin-resistant Staphylococcus aureus (MRSA), a drug-resistant bacterium that causes over 10,000 deaths annually in the US, according to a study published in Nature. Utilising deep learning, a form of AI, the researchers were able to identify features that would enable the compounds to function as potential antibiotics, giving a level of insight not achieved before. The technique also showed low toxicity to human cells, making the compounds a suitable base for future drug candidates.
Leading the study for MIT’s Institute for Medical Engineering and Science (IMES) and the Department of Biological Engineering, James Collins expanded the use of the AI models to serve the Antibiotics–AI Project, which aims to discover new antibiotic classes against seven deadly bacteria types over the next seven years. MRSA infects around 80,000 people each year in the US, with serious cases leading to potentially fatal bloodstream infections.
However, according to Felix Wong, a postdoc at IMES and the Broad Institute, the limitation to the approach is that the AI models are “black boxes”. He explained that the calculations the AI uses are unclear, so researchers developed methods to understand its predictions and aid future antibiotic designs. The researchers trained their deep learning model using data from 39,000 compounds tested for antibiotic activity against MRSA.
Using this method, they were able to screen approximately 12 million commercially available compounds, from which they identified those with potential activity against MRSA. Two of these were found to be particularly promising antibiotic candidates in testing, as they appeared to reduce the MRSA population by a factor of 10 in MRSA skin infection and systemic infection mouse models.
Collins and his team have shared their innovative approach with Phare Bio, a non-profit part of the Antibiotics-AI Project aiming to conduct more detailed analysis on the chemical properties of the compounds and their potential clinical use. The research team is now working on designing additional drug candidates based on their findings and seeking compounds capable of combating other bacteria types using the models developed in the study.
