Using AI, MIT researchers discover a brand new class of antibiotic candidates

Using a kind of artificial intelligence often called deep learning, MIT researchers have discovered a category of compounds that may kill a drug-resistant bacteria that causes greater than 10,000 deaths annually within the United States.

In one Study appears today in , the researchers showed that these compounds can kill methicillin-resistant (MRSA) bacteria grown in a laboratory dish and in two mouse models of MRSA infection. The compounds also exhibit very low toxicity to human cells, making them particularly good drug candidates.

An necessary innovation of the brand new study is that the researchers were also capable of discover what kinds of information the deep learning model used to make its predictions about antibiotic effectiveness. This knowledge could help researchers develop additional drugs that may match even higher than those identified within the model.

“The insight here was that we could see what the models had learned to tell their predictions that certain molecules would make good antibiotics. Our work provides a framework that’s time-efficient, resource-efficient and mechanistically insightful, from a chemical structure standpoint, in a way that we have now not had before,” says James Collins, Termeer Professor of Medical Engineering and Natural Sciences within the Institute for Medical Engineering and Science (IMES) at MIT and within the Department of Biological Engineering.

Felix Wong, a postdoctoral fellow at IMES and the Broad Institute of MIT and Harvard, and Erica Zheng, a former Harvard Medical School graduate student who was advised by Collins, are the lead authors of the study included within the study Antibiotic AI project at MIT. The mission of this project, led by Collins, is to find recent classes of antibiotics against seven kinds of deadly bacteria over a seven-year period.

Explainable predictions

MRSA, which infects greater than 80,000 people within the United States annually, often causes skin infections or pneumonia. In severe cases, it could possibly result in sepsis, a potentially fatal bloodstream infection.

In recent years, Collins and his colleagues at MIT’s Abdul Latif Jameel Clinic for Machine Learning in Health (Jameel Clinic) have begun using deep learning to search out recent antibiotics. Their work has produced potential drugs against the bacteria commonly present in hospitals and plenty of other drug-resistant bacteria.

These compounds were identified using deep learning models that may learn to discover chemical structures related to antimicrobial activity. These models then sift through tens of millions of other compounds and make predictions about which of them might need potent antimicrobial activity.

This kind of search has proven successful, but one limitation of this approach is that the models are “black boxes,” meaning there isn’t a way of knowing on which features the model bases its predictions based. If scientists knew how the models make their predictions, it might be easier for them to discover or develop additional antibiotics.

“Our goal on this study was to open the black box,” says Wong. “These models consist of plenty of calculations that mimic neural connections, and nobody really knows what’s occurring under the hood.”

First, the researchers trained a deep learning model with significantly expanded data sets. They generated this training data by testing about 39,000 compounds for his or her antibiotic activity against MRSA, after which fed that data into the model together with information concerning the compounds’ chemical structures.

“You can principally represent any molecule as a chemical structure and likewise tell the model whether that chemical structure is antibacterial or not,” says Wong. “The model is trained on many examples like this. If you then give it a brand new molecule, a brand new arrangement of atoms and bonds, you may derive a probability that this compound is more likely to be antibacterial.”

To work out how the model made its predictions, the researchers adapted an algorithm often called Monte Carlo tree search, which has been used to make other deep learning models like AlphaGo more explainable. This search algorithm allows the model to supply not only an estimate of every molecule’s antimicrobial activity, but additionally a prediction of which substructures of the molecule are likely accountable for that activity.

Strong activity

To further narrow the pool of drug candidates, the researchers trained three additional deep learning models to predict whether the compounds were toxic to a few various kinds of human cells. By combining this information with predictions of antimicrobial activity, researchers discovered compounds that may kill microbes while causing minimal harmful effects on the human body.

Using this collection of models, the researchers examined roughly 12 million compounds, all of that are commercially available. From this collection, the models identified compounds from five different classes, based on chemical substructures inside the molecules, that were predicted to be lively against MRSA.

The researchers purchased about 280 compounds and tested them against MRSA grown in a lab dish. This allowed them to discover two from the identical class that seemed to be promising antibiotic candidates. In tests on two mouse models, one for MRSA skin infection and one for systemic MRSA infection, each of those compounds reduced the MRSA population by an element of 10.

Experiments found that the compounds appeared to kill bacteria by disrupting their ability to keep up an electrochemical gradient across their cell membranes. This gradient is required for a lot of necessary cellular functions, including the power to supply ATP (molecules that cells use to store energy). An antibiotic candidate that Collins’ lab discovered in 2020, halicin, appears to work by an analogous mechanism but is particular against gram-negative bacteria (bacteria with thin cell partitions). MRSA is a gram-positive bacterium with thicker cell partitions.

“We have fairly strong evidence that this recent class of structures is lively against Gram-positive pathogens by selectively dissipating the proton driving force in bacteria,” says Wong. “The molecules selectively attack bacterial cell membranes in a way that doesn’t significantly damage human cell membranes. Our significantly expanded deep learning approach allowed us to predict this recent structural class of antibiotics and conclude that they’re non-toxic to human cells.”

The researchers shared their findings Phare Bio, a nonprofit founded by Collins and others as a part of the Antibiotics AI Project. The nonprofit organization now plans to conduct a more detailed evaluation of the chemical properties and potential clinical use of those compounds. Meanwhile, Collins’ lab is working to develop additional drug candidates based on the outcomes of the brand new study, using the models to go looking for compounds that may kill other kinds of bacteria.

“We are already using similar approaches based on chemical substructures to develop compounds de novo, and naturally we will immediately adopt this approach to find recent classes of antibiotics against various pathogens,” says Wong.

In addition to MIT, Harvard and the Broad Institute, the institutions involved within the work are Integrated Biosciences, Inc., the Wyss Institute for Biologically Inspired Engineering and the Leibniz Institute for Polymer Research in Dresden, Germany. The research was supported by the James S. McDonnell Foundation, the US National Institute of Allergy and Infectious Diseases, the Swiss National Science Foundation, the Banting Fellowships Program, the Volkswagen Foundation, the Defense Threat Reduction Agency, the US National Institutes of Health, financed. and the Broad Institute. The Antibiotics AI project is funded by the Audacious Project, Flu Lab, the Sea Grape Foundation, the Wyss Foundation and an anonymous donor.