Please check our new paper:
Absorb & Escape: Overcoming Single Model Limitations in Generating Genomic Sequences (Neural Information Processing Systems, 2024) Paper Link
Access the dataset used to replicate the results presented in the paper at Hugging Face: Latent DNA Diffusion Dataset
- Clone the repo
git clone https://github.com/Zehui127/Latent-DNA-Diffusion.git
- Install the required packages: diffusers, accelerate
pip install torch torchvision diffusers accelerate einops easydict pytorch_lightning
A pre-trained vae model and unet model can be downloaded from the following links:
https://zenodo.org/records/11061611
Set the path to the pretrained vae model and unet model in the following command. The number of sequences to generate can be set using the sequence_num argument in src/configs/generate.yaml
.
CUDA_VISIBLE_DEVICES="0" accelerate launch --main_process_port 12903 --multi_gpu main.py --model generate --gen_vae_path="" --gen_unet_path=""
Set the path to the dataset in the config file src/configs/un_unet.yaml
data_path field.
Set the path to the pretrained vae model in the config file src/configs/un_unet.yaml
vae_path field.
CUDA_VISIBLE_DEVICES="0,1" accelerate launch --main_process_port 12903 --multi_gpu main.py --model un_unet
Set the path to the dataset in the config file
CUDA_VISIBLE_DEVICES="0,1" accelerate launch --main_process_port 12903 --multi_gpu main.py --model vanilla_vae
If you use this project or dataset in your research, please cite it as follows:
@article{li2023latent,
title={Latent Diffusion Model for DNA Sequence Generation},
author={Li, Zehui* and Ni, Yuhao* and Huygelen, Tim August B and Das, Akashaditya and Xia, Guoxuan and Stan, Guy-Bart and Zhao, Yiren},
journal={arXiv preprint arXiv:2310.06150},
year={2023}
}
@article{li2024discdiff,
title={DiscDiff: Latent Diffusion Model for DNA Sequence Generation},
author={Li, Zehui and Ni, Yuhao and Beardall, William AV and Xia, Guoxuan and Das, Akashaditya and Huygelen, Tim August B and Stan, Guy-Bart and Zhao, Yiren},
journal={arXiv preprint arXiv:2402.06079},
year={2024}
}