bioRxiv· Xu, Q., Wu, Z., Dong, M., Bao, T., ng, W., Huang, X., Yang, H., Zhou, Y., Yi, X., Zhang, S., Huang, D., Sham, P. C., Zhou, W., Li, M. J. ·· 15 hours agoScore61
A genomic foundation model for interpreting human genetic diversity
In brief
A research team released hEVO, a genomic foundation model pretrained without labels on 4.784 billion haplotype sequences reconstructed from whole-genome sequencing data of 63,809 individuals across eight cohorts. In 19 primary comparisons against models including Evo 2, NT and GPN, hEVO ranked first in 18, and its pretrained allele scores achieved the highest results on all six variant benchmarks. It has fewer parameters than billion-parameter comparators, and a retrospective ClinVar pathogenicity reclassification review identified pathogenic determinations that had later been downgraded.
Source: bioRxiv