The team succeeded in developing a system that could find molecular structures with desired traits (say, kill bacteria) more effectively than systems of the past. Unlike previous methods, neural networks automatically learn the representations of molecules, mapping them into continuous vectors that help predict their behavior. Once ready, the researchers trained their AI on 2,500 molecules that included both 1,700 established drugs and 800 natural products. When asked to examine a library of 6,000 compounds, AI discovered that alicin would be highly effective.
Don’t expect a prescription for halicin soon. MIT has successfully used medicine to eradicate A. baumanii (a common infection for U.S. soldiers in Afghanistan and Iraq) in mice, but did not use it in human trials. This could only be the beginning of a much broader trend, mind you. Scientists have already used their model to select over 100 million molecules in another database, finding 23 candidates. They also hope to design antibiotics from scratch and modify existing drugs to increase their effectiveness or reduce their unwanted side effects. This is far from guaranteed to end up “superbugs”. If he takes away some of them, it could save many lives.