GCNfold: A novel lightweight model for RNA secondary structure prediction

Paper Publication

This article has been published on Computers in Biology and Medicine (SCI Q2 Area, IF: 7.70). You can access it in 50 days (will expire on Sep 13, 2023) via the download link.

Prepare for Experiments

Let us spend some time configuring the environment that GCNfold needs.

Install RNAplfold

RNAplfold is an executable file in the ViennaRNA package that generates the pairing probability of each base in the RNA sequence. We need to download the installation package from ViennaRNA website, tape archive the .tar.gz file, and set the permission of the parent directory.

mkdir rnafold && cd rnafold
wget https://www.tbi.univie.ac.at/RNA/download/sourcecode/2_5_x/ViennaRNA-2.5.0.tar.gz
tar xvzf ViennaRNA-2.5.0.tar.gz
chmod 755 -R /root/rnafold  # change to your file path

We can start installing RNAplfold with the command below. Note that the make operation takes 20 minutes. The make process will appear many warnings. Just ignore them. This will not affect our use.

cd ViennaRNA-2.5.0
./configure
make
sudo make install

Run RNAplfold to check whether the installation is successful. You can also find them in /usr/local/bin.

Configure conda environment

We provide environment.yml files including all the environments that GCNfold depends on.

conda env create -f environment.yml
source activate rna_ss
pip install -e .  # install GCNfold module

Some additional packages need to be installed via the pip command. We used the Tsinghua mirror source when installing the dgl module. Not sure if this is possible outside of China. However, installing the 0.4.2 version of dgl by other ways should also be suitable.

pip install forgi
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple/dgl-cu100/ package/dgl_cu100-0.4.2-cp37-cp37m-manylinux1_x86_64.whl

Data Download

All data used in the experiments have been shared to Google Drive.

ArchiveII dataset testing

ArchiveII is a dataset for extrapolation prediction. You can download it via Google Drive link. Please put it in the folder data/archiveII_all. Run the command below to test. The performance results will be printed on the screen. RNA secondary structures will be stored in the ct_file folder in .ct file format. You can go through RNApdbee to visualize them.

python origin_test.py -c configs/archiveii_test.json

DIY dataset testing

We also support you in building your own datasets. Please put your data (only .ct files are supported) in the data/raw_data/diy_data folder. We have stored 10 RNA data under this path in advance for your testing. Then go through the command below to package them into a .pickle file.

python preprocess_diy_data.py

Perhaps you need to remove duplicate data. This will generate a file called test_no_redundant.pkl under the data/diy_data file.

python filter_redundant_diy_data.py

Preparation is complete. Go ahead and test your DIY dataset below.

python diy_data_test.py -c configs/diy_data_test.json

Citation

APA format:

Yang, E., Zhang, H., Zang, Z., Zhou, Z., Wang, S., Liu, Z., & Liu, Y. (2023). GCNfold: A novel lightweight model with valid extractors for RNA secondary structure prediction. Computers in Biology and Medicine, 107246.

BibTex format:

@article{yang2023gcnfold,
  title={GCNfold: A novel lightweight model with valid extractors for RNA secondary structure prediction},
  author={Yang, Enbin and Zhang, Hao and Zang, Zinan and Zhou, Zhiyong and Wang, Shuo and Liu, Zhen and Liu, Yuanning},
  journal={Computers in Biology and Medicine},
  pages={107246},
  year={2023},
  publisher={Elsevier}
}