hongliangduan / pepscaf

PepScaf

This repository contains the implementation code for paper PepScaf: Harnessing Machine Learning with In Vitro Selection toward De Novo Macrocyclic Peptides against IL-17C/IL-17RE Interaction.

In this study, we first conducted the in vitro selection of macrocyclic peptides against interleukin-17C (IL-17C) using a primary library (17C-Lib1), which gave us a macrocyclic peptide ligand capable of inhibiting IL-17C/IL-17RE interaction with the $IC_{50}$ value at 166 nM. To further improve the activity, we built a framework termed PepScaf to generate the critical scaffold relative to the bioactivity on the basis of the vast dataset generated from the 4th round of 17C-Lib1. Based on the generated scaffold, a focus library (17C-Lib2) was constructed and applied in a macrocyclic peptide selection against IL-17C again. This afforded us with 20 biologically active macrocyclic peptides against IL-17C/IL-17RE interaction with $IC_{50}$ values below 10 nM, of which the best two macrocyclic peptides exhibited their notable inhibitory activities with both IC50 values at 1.4 nM.

Installation

The code was test on GPU 3060 with

• python=3.8
• pytorch==1.12.1
• cudatoolkit=11.3

Conda

We use conda to install the dependencies for PepScaf from the provided environment.yml file, which can give you the exact python environment we run the code for the paper:

NOTE: we also highly recommend using mamba instead of vanilla conda for managing your conda environments. Mamba is a drop-in replacement for conda that is:

• Faster at solving environments (>10x in my experience)
• Better at resolving conflicts
• More informative when something goes wrong.

git clone https://github.com/hongliangduan/PepScaf.git
cd PepScaf
mamba env create -f environment.yml
pip install sci-ztools==0.1.0

mamba activate pepscaf

Usage

Preprocess

python bin/preprocessing.py  # preprocessing for raw data
python bin/cluster.py  # get cluster to get targets

Train and Evaluation

bash train.sh
bash eval.sh

Analysis and Gneration

bash get_attn.sh

Run notebooks/Score.ipynb to get position scores.

bash mcts.sh  # run mcts to get the scaffold

Acknowledgements

1. The implementation of Pep-BERT is partly inspired by A transformer-based model to predict peptide–HLA class I binding and optimize mutated peptides for vaccine design and MolSearch: Search-based multi-objective molecular generation and property optimization.
2. The Pep-BERT was build based fon pytorchic-bert, which is a re-implementation of Google BERT model in Pytorch.
3. The code was formattered by Black.
4. The building of MCTS parts refered to int8's mcts.
5. Weblog
6. The visualization of attention was modified from Pytorch Community
7. The CD-HIT tool and Biopython was used for clustering sequence-based macrocyclic peptides.

Citation

If you find the idea or code useful for your research, please cite our paper:

@article{doi:10.1021/acs.jmedchem.3c00627,
author = {Zhai, Silong and Tan, Yahong and Zhang, Chengyun and Hipolito, Christopher John and Song, Lulu and Zhu, Cheng and Zhang, Youming and Duan, Hongliang and Yin, Yizhen},
title = {PepScaf: Harnessing Machine Learning with In Vitro Selection toward De Novo Macrocyclic Peptides against IL-17C/IL-17RE Interaction},
journal = {Journal of Medicinal Chemistry},
volume = {0},
number = {0},
pages = {null},
year = {0},
doi = {10.1021/acs.jmedchem.3c00627},
    note ={PMID: 37480587},

URL = { 
        https://doi.org/10.1021/acs.jmedchem.3c00627
    
},
eprint = { 
        https://doi.org/10.1021/acs.jmedchem.3c00627
    
}

Contact

Please contact [email protected] if you have any question. Enjoy!