Washington: Every human has a genome that is complex. That’s the equivalent of three billion letters of code, with multiple variations within each individual. It’s probably not humanly possible to sit down and analyze all that code.
However, according to A Research Published in the journal “Nature Machine Intelligence”, artificial intelligence Billions have the ability to detect things that humans might miss through coding. Someday, even AI-powered genome readers will be able to predict the occurrence of diseases From cancer to the common cold. Unfortunately, AI’s recent surge in popularity has created a bottleneck in innovation.
“It’s like the Wild West right now. Everyone’s doing whatever they want,” says Peter Koo, an assistant professor at Cold Spring Harbor Laboratory (CSHL). Just as Frankenstein’s monster was a mixture of different parts, AI researchers are constantly creating new algorithms from different sources. And it is difficult to judge whether their creations will be good or bad. After all, how can scientists judge “good” and “bad” when dealing with computations beyond human capabilities?
That’s where Gopher, Koo Lab’s newest invention comes in. Gopher (short for Genomic Profile-Model Comprehensive Assessment) is a new method that helps researchers identify the most efficient AI programs for analyzing genomes. “We’ve created a framework where you can compare algorithms more systematically,” explains Zhiqi Tang, a graduate student in Koo’s lab.
Gopher judges AI programs on several criteria: how well they learn the biology of our genome, how accurately they predict important patterns and features, their ability to control background noise, and how interpretable their decisions are. “AI is these powerful algorithms that solve problems for us,” Tang says. But, she notes: “One of the main problems with them is that we don’t know how they came up with these answers.”
GOPHER helped Koo and his team discover the parts of AI algorithms that drive reliability, performance and accuracy. The findings help define key building blocks for creating the most efficient AI algorithms going forward. “We hope this will help people in the future who are new to the field,” says Shushan Tonian, another graduate student in the Koo lab.
Imagine feeling uncomfortable and being able to pinpoint exactly what’s wrong with the push of a button. AI may someday turn this science-fiction trope into a staple of every doctor’s office. Like video-streaming algorithms that learn users’ preferences based on their viewing history, AI programs can identify unique features of our genome that lead to personalized medicine and treatment. The Koo team hopes that GOPHER will help optimize such AI algorithms so that we can trust that they are learning the right things for the right reasons. Tonian says: “If the algorithm is making predictions for the wrong reasons, it’s not going to be helpful.”
