AI Creates Killer Drug
Researchers from Canada and the United States used deep learning techniques to develop an antibiotic that could attack a microbe resistant to antibiotics, acinetobacterbaumannii. This bacteria can cause infection and pneumonia in wounds. BBC reports that a Nature Chemical Biology paper describes how researchers used training data to measure the action of known drugs on the bacteria. The algorithm projected the effects of 6,680 compounds without any data on their effectiveness.
The program was able to reduce the list of candidates in an hour-and-a-half. Nine of the compounds were found to be effective, and one compound, abaucin (now called), was very potent. Although 240 compounds may seem like a lot to test, this is much better than 6,700.
It’s interesting that the new antibiotic appears to only be effective against the microbe in question, which is an advantage. The drug is not available to the public yet, and it may take some time before that happens – due to drug testing. This is a good example of how machine-learning can enhance human brainpower and let scientists and others focus their attention on the important things.
Source:
