AI system predicts medicine's hidden powers

By Mason Inman Treatments for new or drug-resistant infectious diseases may already be in our medicine cabinets, say the molecular biologists responsible for developing an artificial-intelligence system that can predict unknown antibiotic properties of existing drugs. The hope is that the work will result in an armoury of new treatments that can be rushed into service when standard treatments stop being effective or new pathogens arise. “In the case of new infectious diseases, there might be no time to develop a completely new drug from the ground up,” says Artem Cherkasov of the University of British Columbia, in Vancouver, Canada, who made the proposal this week at a meeting of the American Chemical Society in Boston, Massachusetts. However, if the new AI system suggests an existing drug might be an effective antibiotic, it could be quickly tested for efficacy, and then pushed into service, Cherkasov says. And because these drugs would have already been approved for use in people, they wouldn’t have to go through all the clinical trials and lengthy regulatory approvals required of brand-new drugs. Many drugs have unexpected or secondary effects – most famously Viagra, Cherkasov points out. Originally developed to combat high blood pressure and angina, Viagra turned out to be effective at treating erectile dysfunction in men. More mundanely, the commonly used cholesterol-lowering drug lovastatin also acts as a potent antibiotic. Rather than waiting for these extra effects to show up by chance, pharmaceutical companies are now using computers to try to determine what other uses their approved drugs might be useful for. Cherkasov is using the same approach, but to a different end. The drugs wouldn’t have to work perfectly, but just well enough to help people in emergency situations when a new disease shows up. “Finding an existing, well-studied therapeutic agent that will kill an emerging bug might provide a rapid first line of defence,” Cherkasov says. He and his team have created an artificial-intelligence system that can tell an antibiotic drug from one that’s not with high accuracy. They first trained the system by feeding in the structures of thousands of known antibiotics, as well as drugs without any bacteria-killing abilities. The system then uses this information to examine the structure of drugs it hasn’t seen before and predict whether they will kill bacteria or not. Cherkasov says the system throws up all kinds of unexpected results. When the system identifies a potential antibiotic, there is often no way of knowing how it actually works. “The chemical structures of compounds we identify usually look nothing like known antibiotics,” he adds. “But we don’t really care how it works. We just need some first line of defence.” Researchers could do this computational work in preparation for new diseases which would give them a database of candidate antibiotics. Then when a new disease emerges, the drugs could be quickly put to the test, to see if they kill the microbes, he suggests. It is plausible that this approach could work, says Alex Tropsha of the University of North Carolina at Chapel Hill, US. “The general methodology has been around for a while,” Tropsha says. But by applying it to emerging diseases,
  • 首页
  • 游艇租赁
  • 电话
  • 关于我们