Imagine a tiny robot detective, trained on millions of clues, scouring a vast treasure trove for a hidden gem. That’s essentially what MIT researchers did, harnessing the power of AI to unearth a brand new class of antibiotics!
This AI sleuth sniffed out compounds that can slay MRSA, a notorious superbug responsible for over 10,000 deaths in the US alone. Not only do these warriors zap the bad guys, but they’re gentle on human cells, making them promising drug candidates.
But here’s the coolest part: the researchers cracked the AI’s code, figuring out what features it used to spot antibiotic gems. This newfound knowledge is like a treasure map, guiding them towards even more potent weapons against other drug-resistant foes! This breakthrough is a major win in the fight against superbugs, offering fresh hope for millions battling these tenacious infections. With further research, these AI-discovered antibiotics could soon be saving lives on the frontlines of healthcare.
MIT researchers have made a groundbreaking discovery in the fight against antibiotic-resistant bacteria, specifically targeting the deadly methicillin-resistant Staphylococcus aureus (MRSA). Using the power of artificial intelligence (AI), the team identified a new class of compounds capable of killing this notorious bacterium responsible for over 10,000 deaths annually in the United States.
Published in the prestigious journal Nature, the study showcases the compounds’ ability to effectively eliminate MRSA in both laboratory dishes and two mouse models of MRSA infection. Notably, these compounds exhibit minimal toxicity against human cells, positioning them as promising candidates for drug development.
What sets this research apart is the elucidation of the AI model’s decision-making process. Researchers, led by James Collins, the Termeer Professor of Medical Engineering and Science at MIT, successfully unveiled the information the deep-learning model employed to predict antibiotic potency. This newfound knowledge provides a valuable framework, allowing for time-efficient, resource-efficient, and mechanistically insightful drug development from a chemical-structure perspective.
The Antibiotics-AI Project at MIT, spearheaded by Collins, aims to discover new classes of antibiotics targeting seven deadly bacteria types over seven years. MRSA, a significant public health concern infecting over 80,000 people annually in the U.S., was a primary focus.
Utilizing deep learning models, MIT researchers identified chemical structures associated with antimicrobial activity, sifting through millions of compounds. However, a limitation arose from the models being “black boxes,” concealing the features influencing predictions. In response, the team embarked on a mission to “open the black box.” They expanded datasets, training a deep learning model on about 39,000 compounds tested for antibiotic activity against MRSA.
To demystify the model’s predictions, an algorithm called Monte Carlo tree search was adapted, providing estimates of each molecule’s antimicrobial activity and predicting contributing substructures. The researchers then trained three additional deep learning models to assess compound toxicity to different human cell types. Combining this information with antimicrobial predictions, the team identified compounds capable of killing microbes with minimal harm to human cells.
In a screening of 12 million commercially available compounds, the models pinpointed five classes predicted to be active against MRSA based on chemical substructures. After testing around 280 compounds, two emerged as highly promising antibiotic candidates within the same class.
Experiments revealed these compounds disrupt bacteria by affecting their ability to maintain an electrochemical gradient across cell membranes, crucial for vital cell functions. This mechanism is similar to that of halicin, an antibiotic candidate discovered in 2020, but with specificity to Gram-positive bacteria like MRSA.
The researchers shared their findings with Phare Bio, a nonprofit associated with the Antibiotics-AI Project, for in-depth analysis of the compounds’ chemical properties and potential clinical applications. Simultaneously, Collins’ lab is actively designing additional drug candidates and exploring the AI models’ potential in identifying compounds effective against various bacteria types.