A genomic foundation model for interpreting human genetic diversity
- 1Guangzhou Women and Children's Medical Center, Guangzhou Medical University, Guangzhou, China; Department of Bioinformatics, The Province and Ministry Co-sponso
Abstract
Interpreting human genomes requires models that recognize shared biological functions while accounting for variation underlying traits, disease susceptibility and adaptation. Here we present hEVO, a genomic foundation model trained on 4.784 billion distinct haplotype sequences reconstructed from whole genome sequencing (WGS) data for 63,809 individuals selected from 550,237 samples across eight cohorts. Pretraining integrates shared sequence features and naturally occurring allelic differences without clinical or functional labels. Against representatives of the Evo 2, NT, GPN and other genomic model families, hEVO ranked first in 18 of 19 primary comparisons spanning functional annotation, variant interpretation and human evolution. Its pretrained allele scores generalized across all six complementary variant benchmarks without task-specific training, achieving the highest performance across clinical pathogenicity, fine-mapped causal variants and experimentally measured regulatory effects. Retrospective ClinVar reclassification further demonstrated improved identification of pathogenic classifications later downgraded, supporting reassessment of earlier clinical interpretations. Predictive strengths extended to ancestry inference, human accelerated sequences and preference for reported selected alleles. This breadth was achieved with substantially fewer parameters than billion-parameter comparators and low inference overhead in the evaluated workloads. These findings establish human haplotype diversity as a source of transferable biological information for interpreting both the disruption of genome function and the variation retained during human evolution.
Related
- Interpreting Short Tandem Repeat Variations with Plant Genomic Foundation Models in Olive
- Uncovering the historic genetic diversity of Ugandan Robusta coffee germplasm: a foundation for core collection design
- Connectivity, genetic diversity and adaptive potential of an endangered beetle: genomic evidence for veteran tree management
- Historical and Contemporary Management of Genetic Diversity in French Local Horse Breeds: Insights from Genomic Data
- Population Genomic Assessment of Genetic Diversity and Structure in Mackerel Scad (Decapterus macarellus) in the South China Sea Based on RAD-seq.
BibTeX
@misc{xu2026genomic,
title = {{A genomic foundation model for interpreting human genetic diversity}},
author = {Q. Xu and Z. Wu and M. Dong and T. Bao and W. ng and X. Huang and H. Yang and Y. Zhou and X. Yi and S. Zhang and D. Huang and P. C. Sham and W. Zhou and M. J. Li},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.19.760276},
url = {https://www.biorxiv.org/content/10.64898/2026.10.19.760276v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - A genomic foundation model for interpreting human genetic diversity
AU - Q. Xu
AU - Z. Wu
AU - M. Dong
AU - T. Bao
AU - W. ng
AU - X. Huang
AU - H. Yang
AU - Y. Zhou
AU - X. Yi
AU - S. Zhang
AU - D. Huang
AU - P. C. Sham
AU - W. Zhou
AU - M. J. Li
PY - 2026
DA - 2026/10/19
PB - bioRxiv
DO - 10.64898/2026.10.19.760276
UR - https://www.biorxiv.org/content/10.64898/2026.10.19.760276v1
AB - Interpreting human genomes requires models that recognize shared biological functions while accounting for variation underlying traits, disease susceptibility and adaptation. Here we present hEVO, a genomic foundation model trained on 4.784 billion distinct haplotype sequences reconstructed from whole genome sequencing (WGS) data for 63,809 individuals selected from 550,237 samples across eight cohorts. Pretraining integrates shared sequence features and naturally occurring allelic differences without clinical or functional labels. Against representatives of the Evo 2, NT, GPN and other genomic model families, hEVO ranked first in 18 of 19 primary comparisons spanning functional annotation, variant interpretation and human evolution. Its pretrained allele scores generalized across all six complementary variant benchmarks without task-specific training, achieving the highest performance across clinical pathogenicity, fine-mapped causal variants and experimentally measured regulatory effects. Retrospective ClinVar reclassification further demonstrated improved identification of pathogenic classifications later downgraded, supporting reassessment of earlier clinical interpretations. Predictive strengths extended to ancestry inference, human accelerated sequences and preference for reported selected alleles. This breadth was achieved with substantially fewer parameters than billion-parameter comparators and low inference overhead in the evaluated workloads. These findings establish human haplotype diversity as a source of transferable biological information for interpreting both the disruption of genome function and the variation retained during human evolution.
ER -