TL;DR:
- MIT researchers employ deep learning to identify new antibiotic candidates targeting drug-resistant bacteria.
- The study showcases compounds effective against MRSA with low toxicity to human cells.
- The innovative approach reveals insights into the deep learning model’s decision-making process.
- Compounds disrupt bacterial cell membranes, offering a potential solution to antibiotic-resistant infections.
- Nonprofit Phare Bio plans further analysis of these compounds for clinical use.
Main AI News:
In the quest to combat drug-resistant bacteria, MIT researchers have harnessed the power of deep learning, unveiling a groundbreaking class of antibiotic candidates. These promising compounds have the potential to combat a drug-resistant bacterium responsible for more than 10,000 annual fatalities in the United States.
Published in the prestigious journal Nature, the study showcases how these compounds effectively eliminate methicillin-resistant Staphylococcus aureus (MRSA) in laboratory settings and in two distinct mouse models of MRSA infection. Notably, these compounds demonstrate remarkably low toxicity toward human cells, positioning them as exceptionally promising drug candidates.
A Notable Innovation: Deconstructing Deep Learning Insights
One of the remarkable achievements of this study lies in the researchers’ ability to unveil the inner workings of their deep-learning model, shedding light on the critical information it utilizes to predict antibiotic potency. This newfound knowledge holds the potential to revolutionize drug design and development, paving the way for more effective antibiotics.
James Collins, the Termeer Professor of Medical Engineering and Science at MIT’s Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering, explains, “The insight here was that we could see what was being learned by the models to make their predictions that certain molecules would make for good antibiotics. Our work provides a framework that is time-efficient, resource-efficient, and mechanistically insightful, from a chemical-structure standpoint, in ways that we haven’t had to date.”
Leading the Charge: Felix Wong and Erica Zheng
Felix Wong, a postdoc at IMES and the Broad Institute of MIT and Harvard, and Erica Zheng, a former Harvard Medical School graduate student, stand at the forefront of this groundbreaking research. They serve as the lead authors of this study, a pivotal component of the Antibiotics-AI Project at MIT. Spearheaded by James Collins, this initiative aims to uncover new classes of antibiotics targeting seven deadly bacterial strains over a seven-year period.
The MRSA Challenge
MRSA, infecting over 80,000 individuals annually in the United States, is a formidable threat, often causing skin infections, pneumonia, and even life-threatening sepsis. Addressing this issue, Collins and his colleagues at MIT’s Abdul Latif Jameel Clinic for Machine Learning in Health (Jameel Clinic) have employed deep learning techniques to identify novel antibiotics. This endeavor has yielded potential drug candidates against Acinetobacter baumannii, a bacterium commonly found in healthcare settings, as well as other drug-resistant bacteria.
The Black Box Conundrum
While their approach has yielded fruitful results, one significant challenge is the opacity of deep learning models, often referred to as “black boxes.” Scientists face a dilemma as they lack insight into the specific features driving the models’ predictions. To overcome this obstacle, the researchers embarked on a mission to unlock the black box.
Revealing the Model’s Inner Workings
In this study, the researchers embarked on a novel approach, leveraging a deep learning model trained on an expanded dataset. By testing approximately 39,000 compounds for antibiotic activity against MRSA and incorporating chemical structure information, they provided the model with crucial insights.
Wong elaborates, “You can represent basically any molecule as a chemical structure, and also you tell the model if that chemical structure is antibacterial or not. The model is trained on many examples like this. If you then give it any new molecule, a new arrangement of atoms and bonds, it can tell you a probability that that compound is predicted to be antibacterial.“
Unveiling Insights with Monte Carlo Tree Search
To decode the model’s decision-making process, the researchers employed an algorithm known as Monte Carlo tree search. This algorithm not only estimates the antimicrobial activity of each molecule but also predicts the substructures responsible for that activity.
A Promising Discovery: Compounds with Minimal Toxicity
To further refine their selection of candidate drugs, the researchers employed three additional deep learning models to assess the toxicity of these compounds on various human cell types. By integrating this data with antimicrobial predictions, they identified compounds capable of eliminating microbes while inflicting minimal harm on the human body.
Analyzing a Vast Pool: Identifying MRSA-Active Compounds
Employing this arsenal of models, the researchers screened a vast collection of approximately 12 million commercially available compounds. From this extensive pool, the models pinpointed compounds from five distinct classes, defined by unique chemical substructures within the molecules, with predicted activity against MRSA.
A Promising Duo: Fighting MRSA
Subsequent laboratory tests against MRSA cultures revealed two exceptionally promising antibiotic candidates from the same class. In experiments involving two distinct mouse models—one for MRSA skin infection and another for MRSA systemic infection—each of these compounds reduced MRSA populations tenfold.
The Mechanism Unveiled
The compounds appear to combat bacteria by disrupting their ability to maintain an electrochemical gradient across their cell membranes—a critical function for vital cell processes, including ATP production. This mechanism is reminiscent of another antibiotic candidate, halicin, discovered in 2020, which targets Gram-negative bacteria. Notably, MRSA belongs to the Gram-positive bacterial group, characterized by thicker cell walls.
Felix Wong emphasizes, “We have pretty strong evidence that this new structural class is active against Gram-positive pathogens by selectively dissipating the proton motive force in bacteria. The molecules are attacking bacterial cell membranes selectively, in a way that does not incur substantial damage in human cell membranes. Our substantially augmented deep learning approach allowed us to predict this new structural class of antibiotics and enabled the finding that it is not toxic against human cells.”
The Path Forward
The researchers have shared their findings with Phare Bio, a nonprofit founded by Collins and others as part of the Antibiotics-AI Project. This organization intends to conduct in-depth analyses of the chemical properties and potential clinical applications of these compounds. Meanwhile, Collins’ lab is actively designing additional drug candidates based on the insights gleaned from this groundbreaking study, as well as employing their models to seek compounds capable of combating other bacterial threats.
Felix Wong concludes, “We are already leveraging similar approaches based on chemical substructures to design compounds de novo, and of course, we can readily adopt this approach out of the box to discover new classes of antibiotics against different pathogens.“
Conclusion:
MIT’s breakthrough in AI-powered antibiotic discovery offers promising prospects for addressing drug-resistant bacterial infections. The unveiling of a new class of antibiotics, backed by explainable deep learning insights, holds significant potential for the healthcare market. These findings could revolutionize antibiotic development and combat a range of deadly bacterial threats, marking a promising advancement in the field.
