SNS Paper - Flashzoi Models
Collection
Flashzoi models from "Reverse engineering the genomic encodings of the pluripotent epigenome" (Casimov et al.). Borzoi architecture trained from scrat • 10 items • Updated
Flashzoi RF 256k — Trained from Scratch
Borzoi architecture trained from random initialization on mouse ESC-derived CUT&Tag H3K27me3 and H3K4me3 tracks.
- Receptive field: 256k
- Architecture: Borzoi (from scratch, model name "flashzoi")
- Resolution: 32 bp
- Training data: Normalized CUT&Tag from mouse ESC-derived embryoid bodies
- Paper: Casimov, Lifshitz et al., "Reverse engineering the genomic encodings of the pluripotent epigenome"
Usage
torchrun --nproc_per_node=8 infer_borzoi_pytorch.py \
--config config.yaml \
--checkpoint model.safetensors \
--genome_fasta mm10.fa \
--output_dir predictions/
See the paper companion repository for full code and configs.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support