Dark matter in AI genome analysis
Artificial intelligence is now part of our everyday lives. First it was ChatGPT. AI-generated beer and pizza commercials are now available. We can’t always trust AI, but we also can’t always trust ourselves.
Peter Koo, assistant professor at Cold Spring Harbor Laboratory, has discovered that scientists who use popular computational tools for AI predictions pick up too much noise or extra information when analyzing DNA. He’s discovered a way to fix it. With just a few lines of new code, scientists are able to get better explanations from powerful AIs called deep neural networks. This allows them to continue searching for real DNA features. These features could be the next big breakthrough in medicine and health. Scientists won’t be able to see signals if the noise is too loud.
What is the source of this obtrusive noise? It is a mysterious source, like digital \”dark material.\” Astronomers and physicists believe that the majority of the universe contains dark matter. This is a material which exerts gravitational forces but has not yet been seen. Koo’s team also discovered that the data used to train AI was lacking critical information. This led to blind spots. These blind spots are also taken into account when AI predicts DNA function. The study was published in Genome Biology.
Source:
https://phys.org/news/2023-06-digital-dark-clouding-ai-genome.html
