Researchers from MIT have utilized deep learning, a form of artificial intelligence, to find a class of compounds that can kill drug-resistant bacteria, specifically methicillin-resistant Staphylococcus aureus (MRSA). The significance of their research is that these compounds have low toxicity against human cells, making them suitable candidates for therapeutic drugs.
Crucially, the researchers can understand the information that the deep learning model is utilizing to make predictions, which can then be used to create more effective drugs. The deep learning method provided a comprehensive overview that allowed the researchers to identify the chemicals that would make effective antibiotics in an efficient and resourceful way.
These compounds were identified using deep learning models to recognize chemical structures that are associated with antimicrobial activity, which is followed by scanning millions of other compounds to predict which ones may have strong antimicrobial activity. An issue with this method is being unaware of what features the model used for its predictions. The researchers have found a way to understand the underlying processes of their modeling through a concept known as the Monte Carlo tree search algorithm.
The researchers tested around 39,000 compounds against MRSA to gain data that was then inputted into the model. The model could then make a prediction on the antibacterial quality of any new molecule. In the antimicrobial activity prediction, different substructures of the molecule were also estimated for their contribution to the activity.
To pinpoint the best drug candidates, additional deep learning models were used to predict toxic effects on different types of human cells. After this, researchers examined 12 million commercially available compounds to identify those predicted to be active against MRSA. From this pool, the researchers identified 280 compounds and tested them against MRSA in a lab setting, eventually finding two that showed promise.
The two potent compounds appear to kill bacteria by disrupting their ability to maintain an electrochemical gradient across their cell membranes. They attack bacterial cell membranes selectively, in a way that does not significantly damage human cell membranes. Phare Bio, a nonprofit founded by Collins and others, will undertake more detailed analyses of the compound’s chemical properties and potential clinical use. The MIT lab is also focusing on designing additional drug candidates based on these discoveries and seeking compounds that can kill other types of bacteria.
The findings are a major step forward in creating new classes of antibiotics to combat drug-resistant bacteria, leading to potentially more effective and efficient treatments.
