CLAPE Framework

Contact: For any questions regarding the code or data, please contact Yufan Liu at andyalbert97@gmail.com.

This repository contains the code for the CLAPE (Contrastive Learning And Pre-trained Encoder) framework for protein–ligand binding site prediction. We provide three ligand-binding tasks: protein–DNA, protein–RNA, and antibody–antigen binding site prediction. For support of small molecule binding site weights and training scripts for the CLAPE series of models, please refer to CLAPE-SMB.

---

Usage

Note: For comprehensive usage instructions, please refer to our CLAPE protocol in Methods in Molecular Biology.

CLAPE primarily depends on the large-scale pre-trained protein language model ProtBert, implemented using HuggingFace Transformers and PyTorch. Please install the dependencies in advance, or create a conda/mamba environment using the provided environment file. If you are using CLAPE-SMB, please also install ESM.

wget https://github.com/YAndrewL/CLAPE/blob/main/environment.yaml
conda env create -f environment.yaml
conda activate clape 

1. Python Package from PyPI

We provide a Python package for predicting ligand-binding sites from protein sequences in FASTA format. Below is an example for DNA-binding site prediction:

# Download model weights and example file
wget https://github.com/YAndrewL/CLAPE/blob/main/example.fa
wget https://github.com/YAndrewL/CLAPE/blob/main/weights/DNA.pth
pip install clape  # Install CLAPE from PyPI

# Example usage
from clape import Clape

model = Clape(model_path="model_path", ligand="DNA")
results = model.predict(input_file="example.fa")

You can set keep_score=True to retain the predicted scores from the model, and use switch_ligand to change the binding site prediction task.

2. Command Line Tools

A command line tool is also provided and installed with the Python package:

clape --input example.fa --output out.txt --ligand DNA --model /path/to/downloaded/model

This command loads the pre-trained models. You can specify the download directory using the --cache parameter.

Parameter descriptions:

| Parameter | Description | |-------------|--------------------------------------------------------------------------------------------------| | --help | Show the help documentation. | | --ligand | Specify the ligand for prediction: DNA, RNA, or AB (antibody). | | --threshold | Threshold for identifying binding sites (0–1, default: 0.5). | | --input | Path to the input file in FASTA format. | | --output | Path to the output file. The first two lines match the input; the third line is the prediction. | | --cache | Path for saving pre-trained parameters (default: protbert). | | --model | Path to trained backbone models. |

---

Citation

If you find our work helpful, please cite as follows:

CLAPE:

@article{10.1093/bib/bbad488,
    author = {Liu, Yufan and Tian, Boxue},
    title = "{Protein–DNA binding sites prediction based on pre-trained protein language model and contrastive learning}",
    journal = {Briefings in Bioinformatics},
    volume = {25},
    number = {1},
    pages = {bbad488},
    year = {2024},
    month = {01},
    issn = {1477-4054},
    doi = {10.1093/bib/bbad488},
    url = {https://doi.org/10.1093/bib/bbad488},
    eprint = {https://academic.oup.com/bib/article-pdf/25/1/bbad488/55381199/bbad488.pdf},
}

CLAPE-SMB:

@article{10.1186/s13321-024-00920-2,
    author={Wang, Jue and Liu, Yufan and Tian, Boxue},
    title={Protein-small molecule binding site prediction based on a pre-trained protein language model with contrastive learning},
    journal={Journal of Cheminformatics},
    year={2024},
    month={Nov},
    day={06},
    volume={16},
    number={1},
    pages={125},
    issn={1758-2946},
    doi={10.1186/s13321-024-00920-2},
    url={https://doi.org/10.1186/s13321-024-00920-2}
}

CLAPE protocol:

@inbook{Liu2025,
  author    = {Yufan Liu and Boxue Tian},
  editor    = {Dukka B. KC},
  title     = {CLAPE: Protein--Ligand Binding Site Prediction via Protein Language Models},
  booktitle = {Large Language Models (LLMs) in Protein Bioinformatics},
  year      = {2025},
  publisher = {Springer US},
  address   = {New York, NY},
  pages     = {293--311},
  isbn      = {978-1-0716-4623-6},
  doi       = {10.1007/978-1-0716-4623-6_18},
  url       = {https://doi.org/10.1007/978-1-0716-4623-6_18}
}

---

Updates

• Jul. 2025: CLAPE protocol published in Methods in Molecular Biology. See the online version.
• Nov. 2024: CLAPE-SMB published in Journal of Cheminformatics. See the online version.
• Aug. 2024: CLAPE is now available as a Python package. See clape on PyPI.
• Mar. 2024: Training code released with CLAPE-SMB. See this repo for reference.
• Jan. 2024: Our paper published in Briefings in Bioinformatics. See the online version.