By James Gamble via SWNS
Artificial intelligence has discovered a class of compounds that can kill hospital superbug MRSA.
American researchers used an AI model to search for compounds to tackle the drug-resistant bacterium which can spread through hospitals and is responsible for as many as 120,000 deaths across the globe every year.
In harnessing the robotic intelligence model, scientists were also able to decipher the kinds of information the deep-learning model was used to make its antibiotic potency predictions.
The researchers hope this knowledge could help them design additional drugs that might work even better than the ones identified by the model and could help to save millions of lives across the world.
The compounds also show very low toxicity against human cells, making them particularly good drug candidates.
The study, published in the journal Nature, is part of the Antibiotics-AI Project at the Massachusetts Institute of Technology (MIT), which seeks to discover new classes of antibiotics against seven types of deadly bacteria over seven years.
Methicillin-resistant Staphylococcus aureus, commonly known as MRSA, infects more than 80,000 people in the United States alone every year and can often cause skin infections or pneumonia.
Severe cases can lead to sepsis, a bloodstream infection that can prove fatal.
In recent years, researchers at MIT have begun using deep learning, a type of AI that imitates the way humans gain certain types of knowledge, to try to find new antibiotics.
Their work has so far yielded potential drugs against Acinetobacter baumannii, a bacterium that is often found in hospitals, and many other drug-resistant bacteria.
These compounds were identified using deep learning models that can learn to identify chemical structures associated with antimicrobial activity.
These models then sift through millions of other compounds, generating predictions of which ones may have strong antimicrobial activity.
But though these searches have proven fruitful the models are ‘black boxes’, meaning there’s no way of knowing what features the model based its predictions on, which would make it easier for scientists to identify or design antibiotics.
“What we set out to do in this study was to open the black box,” Dr Felix Wong, a postdoc at the Broad Institute of MIT and Harvard, says.
“These models consist of very large numbers of calculations that mimic neural connections, and no one really knows what’s going on underneath the hood.”
The researchers first trained a deep learning model using substantially expanded datasets.
They generated this training data by testing around 39,000 compounds for antibiotic activity against MRSA, and then fed this data, along with information on the chemical structures of the compounds, into the model.
The researchers adapted an algorithm known as Monte Carlo tree search, which has been used to help make other deep learning models, to find out how the model was making its predictions.
To further narrow down the pool of candidate drugs, the researchers trained three more deep-learning models to predict whether the compounds were toxic to three different types of human cells.
By combining this information with the predictions of antimicrobial activity, the researchers discovered compounds that could kill microbes whilst having minimal adverse effects on the human body.
Using their AI models, the MIT researchers screened around 12 million commercially available compounds and identified those from five different classes that were predicted to be active against MRSA.
They then tested 280 compounds against MRSA grown in a lab dish and identified two that appeared to be very promising candidates.
In tests in two mouse models – one of MRSA skin infection and one of MRSA systemic infection – each of those compounds reduced the MRSA population by a factor of 10.
Experiments revealed the compounds appear to kill bacteria by disrupting their ability to maintain an electrochemical gradient across their cell membranes.
On the significance of the study’s findings, James Collins, a professor of Medical Engineering & Science at MIT, said: “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.”
“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,” Dr Wong added.
“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.”
