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
Install pytorch 1.13.1 with the cuda version that is compatible with your device. The geomstats package does not support torch>=2.0.1 on GPU until Mar.30, 2024. Here we recommend using torch==1.13.1.
# 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}
Note
- Due to an error in the validation script where a symbol was incorrectly written, all methods were inadvertently evaluated using fragment segments during validation (See Issue). This led to an overestimation of performance in FoldX. We sincerely apologize for the inaccuracies reported in the paper, and the corrected version of the paper has addressed this issue.
- We optimized the orientation and translation Flow matching using the method described in Sec. 4.1: Diffusion Conditional Vector Fields from FlowMatching. The model will be available through our lab's platform.
If you want to directly evaluate the peptides, we provide the peptides as codesign_results.tar.gz from our google drive, which consists of 20 samples per protein structure for more stable evaluation, with results given as
| Method | IMP%-S(↑) | Validity(↑) | Novelty(↑) | Diversity |
|---|---|---|---|---|
| PPFlow | 4.04% | 1.00 | 0.99 | 0.67 |
| DiffPP | 3.72% | 0.41 | 0.89 | 0.28 |
Although there was a slight decline in the metrics on foldX, the overall performance still comprehensively surpasses DiffPP.
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.
HDock: For HDock, firstly, libfftw3 is needed for hdock with 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.
Pyrosetta: For pyrosetta, you should first sign up at https://www.pyrosetta.org/downloads. After the authorization of the license, you can install it through
conda config --add channels https://yourauthorizedid:password@conda.rosettacommons.org
conda install pyrosetta
./tools/relax/rosetta_packing.py gives an example of our python interface for rosetta side-chain packing.
FoldX: For FoldX, you should register and log in according to https://foldxsuite.crg.eu/foldx4-academic-licence, download the packages, and copy it to ./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.
ADCP: We provide the available ADFRsuite software in ./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
- TMscore
TMscore
TMscore.cpp
PLIP: If you want to analyze the interaction type of the generated protein-peptide, you can use PLIP: https://github.com/pharmai/plip.
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},
}