ppflow
1.0.0
conda create -n ppflow python==3.9
conda activate ppflow
# install requirements
pip install -r requirements.txt
pip install easydict
pip install biopython
# mmseq
conda install bioconda::mmseqs2
# Alternative: obabel and RDkit
conda install -c openbabel openbabel
conda install conda-forge::rdkit
# Alternative for visualization: py3dmol
conda install conda-forge::py3dmol
# torch-geomstats
conda install -c conda-forge geomstats
# torch-scatter
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.13.0+cu117.html
# OR: stable torch-scatter
pip install ./temp/torch_scatter-2.1.1+pt113cu117-cp39-cp39-linux_x86_64.whl
We provide the processed dataset of PPBench2024 through google drive, together with processed `PPDBench'.
Please download data.zip and unzip it, leading to the data file directory as
- data
- processed
cluster_result_all_seqs.fasta
cluster_result_cluster.tsv
cluster_result_rep_seq.fasta
parsed_pair.pt
receptor_sequences.fasta
split.pt
- processed_bench
cluster_result_all_seqs.fasta
cluster_result_cluster.tsv
cluster_result_rep_seq.fasta
parsed_pair.pt
receptor_sequences.fasta
split.pt
pdb_benchmark.pt
pdb_filtered.pt
If you want the raw datasets for preprocessing, please download them through google drive. Unzip the file of datasets_raw.zip, leading to the directory as
- dataset
- PPDbench
- 1cjr
peptide.pdb
recepotor.pdb
- 1cka
peptide.pdb
recepotor.pdb
...
- ppbench2024
- 1a0m_A
peptide.pdb
recepotor.pdb
Run the following command for PPFlow training:
python train_ppf.py
Run the following command for DiffPP training:
python train_diffpp.py
For RDE finetuning, you should first download the pretrained RDE.pt model from the google drive, then save it as ./pretrained/RDE.pt, and finally, run the following command for finetuning:
python train_rde.py --fine_tune ./pretrained/RDE.pt
python codesign_diffpp.py -ckpt {where-the-trained-ckpt-is}
python codesign_ppflow.py -ckpt {where-the-trained-ckpt-is}
Here we give the checkpoints that are pretrained, which is named ppflow_pretrained.pt and can be downloaded from the google drive. You can directly download it and copy it to ./pretrained/ppflow_pretrained.pt. Further, run the following to generation:
python codesign_diffpp.py -ckpt ./pretrained/ppflow_pretrained.pt
If you want to directly evaluate the peptides, we provide the peptides as codesign_results.tar.gz from our google drive, which consists of 100 samples / protein structure for more stable evaluation, with results given as
| IMP%-S(↑) | Validity(↑) | Novelty(↑) | Diversity |
|---|---|---|---|
| 12.50% | 1.00 | 0.99 | 0.92 |
You should evaluate files that end with _bb3.pdb as the generated pdb, since the O element in _bb4.pdb is unstable in our reconstruction function.
conda install conda-forge::vina
pip install meeko
pip install git+https://github.com/Valdes-Tresanco-MS/AutoDockTools_py3.git@aee55d50d5bdcfdbcd80220499df8cde2a8f4b2a
pip install pdb2pqr
./tools/dock/vinadock.py gives an example of our Python interface for vinadock.
apt-get install -y libfftw3-3. Besides, the HDock software can be downloaded through: http://huanglab.phys.hust.edu.cn/software/hdocklite/. After downloading it, install or unzip it to the ./bin directory, leading to the file structure as- bin
- hdock
1CGl_l_b.pdb
1CGl_r_b.pdb
createpl
hdock
./tools/dock/hdock.py gives an example of our python interface for hdock.
conda config --add channels https://yourauthorizedid:[email protected]
conda install pyrosetta
./tools/relax/rosetta_packing.py gives an example of our python interface for rosetta side-chain packing.
./bin. Then, unzip it, which will lead the directory to look like- bin
- FoldX
foldx
where foldx is the software. ./tools/score/foldx_energy.py gives an example of our Python interface for foldx stability.
./bin. If it is not compatible with your system, please install it through https://ccsb.scripps.edu/adcp/downloads/. Copy the ADFRsuite_x86_64Linux_1.0.tar into ./bin. Finally, the installed ADCP into ./bin should look like- bin
- ADFRsuite_x86_64Linux_1.0
- Tools
CCSBpckgs.tar.gz
...
ADFRsuite_Linux-x86_64_1.0_install.run
uninstall
Remember to add it to your env-path as
export PATH={Absolute-path-of-ppfolw}/bin/ADFRsuite_x86_64Linux_1.0/bin:$PATH
./tools/dock/adcpdock.py gives an example of our Python interface for ADCPDocking.
./bin, as- bin
- TMscore
TMscore
TMscore.cpp
First, clone it to ./bin
cd ./bin
git clone https://github.com/pharmai/plip.git
cd plip
python setup.py install
alias plip='python {Absolute-path-of-ppfolw}/bin/plip/plip/plipcmd.py'
./tools/interaction/interaction_analysis.py gives an example of our Python interface for plip interaction analysis.
The evaluation scripts are given in ./evaluation directory, you can run the following for the evaluation in the main experiments:
# Evaluting the docking energy
python eval_bind.py --gen_dir {where-the-generated-peptide-is} --ref_dir {where-the-protein-pdb-is} --save_path {where-you-want-the-docked-peptide-to-be-saved-in}
# Evaluating the sequence and structure
python eval_bind.py --gen_dir {where-the-generated-peptide-is} --ref_dir {where-the-protein-pdb-is} --save_path {where-you-want-the-docked-peptide-to-be-saved-in}
We give an example file pair, so the gen_dir can be ./results/ppflow/codesign_ppflow/0008_3tzy_2024_01_19__19_16_21, ref_dir can be ./PPDbench/3tzy/ and the save_path can be ./results/ppflow/codesign_ppflow/0008_3tzy_2024_01_19__19_16_21.
If our paper or the code in the repository is helpful to you, please cite the following:
@inproceedings{lin2024ppflow,
author = {Lin, Haitao and Zhang, Odin and Zhao, Huifeng and Jiang, Dejun and Wu, Lirong and Liu, Zicheng and Huang, Yufei and Li, Stan Z.},
title = {PPFlow: Target-Aware Peptide Design with Torsional Flow Matching},
year = {2024},
booktitle={International Conference on Machine Learning},
URL = {https://www.biorxiv.org/content/early/2024/03/08/2024.03.07.583831},
eprint = {https://www.biorxiv.org/content/early/2024/03/08/2024.03.07.583831.full.pdf},
}
