GPN-Star: The AI that learns from evolution to decode the human genome

Researchers at UC Berkeley have developed GPN-Star, an innovative genomic language model for artificial intelligence. This tool excels at identifying the most important genetic variants contributing to heritable traits, including those leading to diseases.

Unlike previous models, GPN-Star was trained using data from whole genome alignments (WGA) rather than unaligned individual genomes. This approach, which relates the genomes of hundreds of different species to that of a reference species, enables the model to more effectively pinpoint areas of conserved code during evolution.

The model offers remarkable computational efficiency, requiring only a fraction of the time and computing resources needed to train larger models. It can be trained in a few days or even hours using only a handful of processors. The researchers have published the model's predictions to guide biological discovery, helping prioritize experiments with the greatest impact on health.