Skip to content

Latest commit

 

History

History
33 lines (23 loc) · 896 Bytes

File metadata and controls

33 lines (23 loc) · 896 Bytes

Efficient Protein Optimization via Structure-aware Hamiltonian Dynamics

Code and data to reproduce results of our work HADES at AAAI 2026: https://arxiv.org/abs/2601.11012

Prepare for conda environment

conda env create -f environment.yml
conda activate hades

(Optional) download predicted structures

download pdb_cache.zip from the following link and unzip:

https://1drv.ms/u/c/D255A9FC0E665D01/IQAWI0xZNg1GTqwUa3GD0uB_AUK1IUsBiifKo_SzVylcpB8

For GB1

python main.py --task GB1 --mut_pos 38 39 40 53

For PhoQ

python main.py --task PhoQ --mut_pos 283 284 287 288

Note

  • check samples files in results folder after training is done
  • --use_structure option runs ESMFold to predict structures during training if you skip step 2, which is time consuming and turned off by default.