Researchers developed GPN-Star, a genomic language model that models evolutionary relationships using species trees and whole-genome alignments. Trained on data spanning the evolutionary periods of vertebrates, mammals and primates, the model outperformed previous methods in predicting variant effects in coding and non-coding regions of the human genome. GPN-Star also delivered strong results in prioritizing variants identified through pathogenicity and genome-wide association studies, explaining complex-trait heritability, and rare-variant association tests. The model was also shown to be applicable to Mus musculus, Gallus gallus, Drosophila melanogaster, Caenorhabditis elegans and Arabidopsis thaliana.
Why it matters
This approach offers a way to assess the effects of variants in genomic research not only by examining individual sequences, but also by leveraging evolutionary relationships between species. This allows researchers to address more systematically which changes, particularly those in non-coding regions and those occurring rarely, may warrant priority for further study. Its applicability to different organisms shows that the method can also be used in comparative genomics studies without being limited to the human genome. However, how the prioritizations produced by the model will translate into interpretations across different biological questions and species remains one of the open questions that will determine the method’s scope of use.
Background
Star is not a new name in the FikirPilot archive: we published a news article mentioning this name in the last 90 days; that article is dated September 12, 2026.