Popformer

An axial attention transformer for haplotype matrices, pre-trained with self-supervised masked haplotype reconstruction.

Paper: Popformer: Learning general signatures of positive selection with a self-supervised transformer

Model Description

Popformer is pre-trained on masked haplotype reconstruction and evaluated on:

  • Natural selection detection
  • Genotype imputation
  • Population classification

Usage

See the repository README for full preprocessing and inference examples, including VCF/HDF5 input and genome-wide selection scans.

Citation

@article{popformer2026,
  title   = {Popformer: Learning general signatures of positive selection with a self-supervised transformer},
  url     = {https://www.biorxiv.org/content/10.64898/2026.03.06.710163v1}
}
Downloads last month
23
Safetensors
Model size
12.9M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support