New AI model for DNA learns from evolution to unlock secrets of the human genome
More than two decades after scientists first sequenced the entire human genome—all 3 billion base pairs, of DNA code—the meaning of much of this code remains a mystery.
Decades after sequencing the entire human genome, scientists still struggle to understand the purpose of most of its noncoding regions. These regions, often referred to as "junk DNA," may hold the answers to diseases like cancer and autism. Researchers at UC Berkeley have developed an AI model called GPN-Star that quickly and efficiently identifies genetic variants that could have a significant impact on human health.
Unlike other models, GPN-Star only requires a fraction of the time and computing resources to train because it uses genome-wide alignments (WGAs) instead of entire unaligned genomes. WGAs identify similarities and differences between species' genomes, allowing GPN-Star to focus on conserved areas of code that are more likely to be functional.
The model was trained on three human-anchored WGAs and on WGAs for other species, which optimized it for interpreting different types of genetic variants. In the future, biologists could use GPN-Star to prioritize experiments that could have the greatest impact on human health.
Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
