PO2Vec is a novel method for GO term representation learning. Compared with existing methods based on GO DAG structure, PO2Vec captures the topological information of GO more comprehensively under in-path partial order constraint and out-path partial order constraint. The effectiveness of PO2Vec was demonstrated in experimental analyses from five aspects.

PO2GO is a two-step protein function annotation prediction model, which is jointly constructed by PO2Vec and the protein language pre-trained model ESM-1b. The superior performance of PO2GO is demonstrated with comparative benchmarks.

This repository contains script which were used to build and train the PO2Vec and PO2GO model.

• The code was developed and tested using python 3.9.
• All the dependencies are listed in requirements.txt file.

The source code for PO2Vec and PO2GO can be obtained from the official GitHub repository.

To clone the repository, use the following command:

# clone project
git clone https://github.com/xbiome/protein-annotation
# install dependencies
pip install -r requirements.txt

Alternatively, you can manually install the necessary dependencies, such as torch via pip.

pip install torch

Once you have a copy of the source code, install it with:

python setup.py install

To use the PO2GO algorithm for more accurate protein function prediction, a model was trained with all annotated proteins from UniProt-SwissProt (approximately 500,000 proteins). The saved model weights and the associated go.obo and terms_annotated_embeddings.pkl files can be accessed from here.

The scripts require GeneOntology in OBO Format.

• preprocess_raw_data.py - This script is used to convert data from UniProt database format to pandas dataframe.
• prepare_data_seperate_swissprot.py - This script is used to seperate data from UniProt-Swissprot into train and test dataset.
• prepare_data_seperate_ontologies.py - This script is used to seperate train(test) data into three groups based on annotation ontologies.
• po2go/po2vec - This folder is used to train PO2Vec model to get terms embeddings.
• train_po2go.py - This script is used to train the model.

• to train a model predict mfo terms run sh:

python train_po2go.py --namespace mfo -ld 512 -p 768

• to train a model predict bpo terms run sh:

python train_po2go.py --namespace bpo -ld 768 -p 1280

• to train a model predict cco terms run sh:

python train_po2go.py --namespace cco -ld 512 -p 896

• to train a model predict annotated terms run sh:

python train_po2go.py --namespace annotated -ld 768 -p 1280

• inference_po2go.py - This script is used to inference test data to get prediction file.
• evaluate_*.py - The scripts are used to compute Fmax, Smin and AUPR.

• inference_fasta.py -This script is used to inference fasta format data by annotated model.

python inference_fasta.py -i input.fasta -o output.csv -b batchsize

python inference_fasta_cpu.py -i input.fasta -o output.csv -b batchsize -T number_cpucore

If you use PO2Vec for your research, or incorporate our learning algorithms in your work, please cite:

@article{li2024partial,
  title={Partial order relation--based gene ontology embedding improves protein function prediction},
  author={Li, Wenjing and Wang, Bin and Dai, Jin and Kou, Yan and Chen, Xiaojun and Pan, Yi and Hu, Shuangwei and Xu, Zhenjiang Zech},
  journal={Briefings in Bioinformatics},
  volume={25},
  number={2},
  pages={bbae077},
  year={2024},
  publisher={Oxford University Press}
}

Current dependencies can be found in the requirements.txt file. The used Python version is 3.9.12.