AIxFuse

Our paper published as a back cover article in Chemical Science:
Structure-aware dual-target drug design through collaborative learning of pharmacophore combination and molecular simulation

Environment

conda env create -f env.yml
, then run conda activate aixfuse
DGLlife with pytorch:
pip3 install dgl -f https://data.dgl.ai/wheels/torch-2.4/cu121/repo.html
, or refer to DGL official to specify your cuda version et al.
, then run pip3 install dgllife
PLIP:
apt install libopenbabel-dev
pip3 install openbabel
, if error occurs like Error: SWIG failed. Is Open Babel installed?:

• run apt install swig
• if it still fail, refer to openbabel issue

, then run pip3 install plip
MOSES:
pip3 install molsets
, if error occurs like ERROR: Failed building wheel for pomegranate:

• run git clone git@github.com:molecularsets/moses.git
• , then run cd moses
• , then run python setup.py install

Download Data (too large for github)

by Google Drive
, or by BaiduNetDisk, fetch code：866p
, or contact My Email

Pharmacophore Extraction

Run pharmacophore_extr.ipynb

Generate Molecules by Our Final Agent

python iter_gen.py --task rorgt_dhodh --iter 4 --single_process --generate --schrodinger_dir xx --ligpre_path xx --glide_path xx
, leave schrodinger_dir, ligpre_path, glide_path as xx, here we don't need those software.

Generate Molecules from Scratch

Make sure that you have schrodinger, ligpre, and glide at your_schrodinger_dir, your_ligpre_path, and your_glide_path.
We run glide on a distributed system. If you don't run it this way, please modify utils/ligpre.py and utils/docking.py.
, then you can run:
python iter_gen.py --task rorgt_dhodh --schrodinger_dir your_schrodinger_dir --ligpre_path your_ligpre_path --glide_path your_glide_path