AllenInstitute/BG-human-dilatedCNN

Dilated CNN trained on human (Homo sapiens) basal ganglia chromatin accessibility data across 59 cell type tracks. Part of the DNA sequence modeling resources associated with the basal ganglia cell atlas package. Published in Johansen, Fu et al., 2025.

Model Details

Field	Value
Architecture	Dilated CNN
Species	Homo sapiens (human)
Brain region	Basal ganglia
Number of tracks	59
Framework	PyTorch Lightning
Checkpoint format	.ckpt

Background

This model learns cis-regulatory sequence logic underlying basal ganglia cell type specialization from snATAC-seq-derived accessibility profiles. It can be used to score candidate enhancer sequences, run in silico mutagenesis, or guide sequence design for specific cell types.

For broader context on the associated multiomic studies and the consensus cell type taxonomy used to define tracks, see the BG Cell Atlas sequence modeling repository.

Availability

This repository is currently private. The enhancer-designer codebase is an internal Allen Institute tool with a public release planned in the near future.

To request early access, please contact Kasia Kedzierska at kasia.kedzierska@alleninstitute.org.

Usage

from enhancerdesigner.models import load_model

model = load_model("AllenInstitute/BG-human-dilatedCNN")

Target Tracks

The full list of 59 tracks is accessible programmatically after loading:

model = load_model("AllenInstitute/BG-human-dilatedCNN")
print(model.target_names)

Citation

@article{Johansen2025,
  title   = {Cross-species consensus atlas of the primate basal ganglia},
  author  = {Johansen, Nelson and Fu, Yuanyuan and others},
  journal = {bioRxiv},
  year    = {2025},
  doi     = {10.64898/2025.12.15.694496}
}