GPN-Star model source code

GPN-Star is a phylogeny-aware genomic language model trained on whole-genome alignments across multiple evolutionary timescales. It is designed for learning genomic sequence constraints and predicting the functional effects of genetic variants. This code-only deposit corresponds to GPN release 0.8 and Git commit 667c353b559916d000ff86867c909ebb334628c5. The archive includes: - The GPN-Star model, training, inference, embedding, and variant-effect-prediction implementation. - Training and evaluation workflows for six species. - S-LDSC workflows, including Borzoi- and cCRE-based analyses. - Whole-genome-alignment processing and model-interpretation workflows. - A GPN-Star example notebook and automated tests. Source repository: https://github.com/songlab-cal/gpn Citation: Ye C, Benegas G, Albors C, Li JC, Prillo S, Fields PD, Clarke B, Song YS. “Predicting functional constraints across evolutionary timescales with phylogeny-informed genomic language models.” bioRxiv (2025). https://doi.org/10.1101/2025.09.21.677619