kaist-silab/bootgen

[NeurIPS 23] Official code for BootGen: "Bootstrapped Training of Score Conditioned Generator for Offline Biological Design"

BootGen

Official code for BootGen: "Bootstrapped Training of Score-Conditioned Generator for Offline Design of Biological Sequences".

Dependencies

• Python==3.7

pip install flexs
conda install -c bioconda viennarna
pip install design-bench==2.0.20
pip install polyleven

Running BootGen

GFP

python train.py --DA --task gfp --lr 5e-5

UTR

python train.py --DA --task utr --lr 5e-5

TFbind 8

python train.py --DA --task tfbind

RNA-A

python train.py --DA --task rna1

RNA-B

python train.py --DA --task rna2

RNA-C

python train.py --DA --task rna3

To test BootGen without "diverse aggregation," simply remove the "--DA" flag.

Cite us

If you find this code useful, please cite our paper:

@article{kim2023bootstrapped,
  title={Bootstrapped Training of Score-Conditioned Generator for Offline Design of Biological Sequences},
  author={Kim, Minsu and Berto, Federico and Ahn, Sungsoo and Park, Jinkyoo},
  journal={arXiv preprint arXiv:2306.03111},
  year={2023}
}