A robotic system developed by a group of Carnegie Mellon scientists could make the drug development process faster, cheaper and more efficient. See, drug discovery, which identifies potential new medicine, entails loads of lab testing to determine the effects of different drugs on their target proteins. So many, in fact, that scientists typically have to choose the experiments to run, else they might never be done.
The study's lead author Armaghan Naik said it can be tough for humans to pick the right experiments: they have to guess the hypothetical outcomes for each one to be able to choose. For their study that used 96 drugs and 96 cells, for instance, there were 9,216 possible experiments. Imagine having to conjure up a hypothetical result for each and every one of those.
Now, here's where the team's algorithm comes in. During their tests, the machine chose a few experiments to run, which were then conducted using liquid-handling robots and automated microscopes. It learned more about the drugs, the cells and how they interact after each round of testing. The AI then used what it learned from the past round to choose what to run in the next. In the end, the machine performed 2,697 out of the 9,216 possible experiments after 30 rounds. The team says the algorithm was "able to learn a 92 percent accurate model for how the 96 drugs affected the 96 proteins" by conducting only 29 percent of the almost 10,000 possible experiments.
The video below shows the tests chosen by the AI each round. But if you're ready and able to grapple with scientific terms and the nitty-gritty of the study, you can check out CMU's longer write up or the paper itself on eLIFE.