Artificial Intelligence (AI) machines can be trained to solve puzzles on their own, by learning to recognise rules and patterns in data, rather than by simply following the rules humans program into them. But often, researchers don’t know what rules the AI have made for themselves.
Peter Koo, an assistant professor at Cold Spring Harbor Laboratory, Long Island, US, has developed a new method – described today in PLOS Computational Biology – that quizzes an AI to figure out what rules it has learned on its own, and whether they’re the right ones.
“If you learn general rules about the math instead of memorizing the equations, you know how to solve those equations. So rather than just memorizing those equations, we hope that these models are learning to solve it and now we can give it any equation and it will solve it,” says Koo.
Koo has developed an AI called a deep neural network (DNN), that looks for patterns in strands of RNA that increase the ability of a protein to bind to them. Koo’s DNN, called Residual Bind (RB), has been trained with thousands of RNA sequences matched to protein binding scores, and is able to predict scores for new RNA sequences.
But Koo was not sure what rules the machine was focusing on – whether it was focusing on a short sequence of RNA letters (a motif) that humans might expect, or whether it was looking at other characteristics. He and his team therefore developed a new method, called Global Importance Analysis, to test what rules RB generated. They presented the network with a set of synthetic RNA sequences containing different combinations of motifs that the scientists thought might influence RB’s calculations.
What they discovered is that RB factored in a range of considerations, including how the RNA strand may fold over and bind to itself, how close one motif is to another, and other features.
With the help of RB and Koo’s new Global Importance Analysis, the team can now test biological results in a ‘virtual’ laboratory, running millions of tests far faster than humans could do in traditional lab settings. Their tools are now available online for anyone to use.
“If you learn general rules about the math instead of memorizing the equations, you know how to solve those equations. So rather than just memorizing those equations, we hope that these models are learning to solve it and now we can give it any equation and it will solve it.”
Amalyah Hart is a science journalist based in Melbourne.
Read science facts, not fiction...
There’s never been a more important time to explain the facts, cherish evidence-based knowledge and to showcase the latest scientific, technological and engineering breakthroughs. Cosmos is published by The Royal Institution of Australia, a charity dedicated to connecting people with the world of science. Financial contributions, however big or small, help us provide access to trusted science information at a time when the world needs it most. Please support us by making a donation or purchasing a subscription today.