Researchers use artificial intelligence to find new antibiotics

Researchers have discovered a new antibiotic using deep learning. Their algorithms predicted based on molecular structure that a drug should work against several bacterial species.

The scientists work at the leading technology institute MIT, and have published their findings in the journal Cell. There they describe setting up a neural network with the aim of finding new antibiotics; this is necessary because many bacterial species become resistant to existing agents.

Because very large numbers of molecules have to be tested for the discovery of new drugs, the researchers have built a prediction model in which it is possible to predict which molecules have an antibacterial effect on the basis of molecular structure. To do this, they trained their neural network with the properties of 2,335 different molecules. The results were then applied to a database of 107 million different molecules.

In their paper, the researchers describe the discovery of the drug halicin, an enzyme inhibitor that was initially developed as a medicine against diabetes. However, the deep learning model predicted that it should also have antibacterial activity based on the molecular structure. After validation of the finding in mice, with a demonstrated effect against resistant bacteria, among other things, this turned out to be the case.

The researchers also found other molecules that may have an antibacterial effect. The findings may eventually lead to new drugs that can be used to combat resistant bacteria.