MIT Leverages AI to Discover New Antibiotics Targeting Superbugs

For years, doctors have faced a formidable challenge in combating bacteria that have developed growing resistance to existing antibiotics. However, researchers from MIT are pioneering a breakthrough using deep learning to identify a new class of compounds that could change the tide in this ongoing battle. In a groundbreaking study published in *Nature*, those researchers discovered compounds capable of effectively killing the notorious MRSA (methicillin-resistant Staphylococcus aureus) in laboratory settings, all while exhibiting low toxicity toward human cells. This promising characteristic positions them as potential candidates for new drugs.

MRSA is responsible for infecting over 80,000 Americans annually, with severe infections leading to serious health complications such as sepsis, a life-threatening condition characterized by poisoning of the blood. Addressing this urgent public health concern, MIT's research team trained a sophisticated deep learning model on robust data generated from testing 39,000 compounds for their antibiotic activity against MRSA. They complemented this dataset with detailed information on the chemical structures of these compounds, enabling the model to pinpoint potential drug candidates.

To refine their search further, the researchers developed three additional deep learning models that predicted which compounds might be toxic to human cells. This multifaceted approach allowed them to identify compounds that could effectively eliminate MRSA without posing risks to human health. Leveraging these AI models, the team then screened a staggering 12 million commercially available compounds, ultimately isolating those belonging to five distinct classes predicted to be effective against MRSA. Following this high-throughput screening, two promising candidates emerged through rigorous laboratory testing.

Leading this innovative study are Felix Wong from MIT and Harvard, along with Erica Zheng, a graduate of Harvard Medical School. Their research is part of the Antibiotics-AI Project at the Collins Lab at MIT, which aims to develop seven new classes of antibiotics targeting some of the world’s deadliest bacteria.

The targeted superbugs include:

- **E. coli**

- **Klebsiella pneumoniae**

- **Acinetobacter baumannii**

- **Pseudomonas aeruginosa**

- **Neisseria gonorrhoeae**

- **Staphylococcus aureus**

- **Mycobacterium tuberculosis**

The lab is committed to creating a comprehensive training library of 100,000 compounds, which will be screened against these seven pathogens to identify active molecules. This invaluable data will serve to further train machine learning models applied to extensive computer-based libraries containing over a billion molecules, ultimately aiming to discover and design novel antibiotics.

According to the Centers for Disease Control, an American is diagnosed with a drug-resistant infection every 11 seconds, with a death occurring every 15 minutes due to such infections. The widespread misuse of antibiotics in both humans and animals has exacerbated this issue, leading to a global crisis in antibiotic resistance. The lab emphasizes the urgent necessity for new antibiotics, highlighting that pharmaceutical companies have largely shifted focus away from this critical area in favor of more profitable markets. The antibiotics being developed through this project would mark the first new classes in over three decades, underscoring the dire need for innovation in this field.