DPFunc

Accurately predicting protein function via deep learning with domain-guided structure information.

🚀 Latest Updates

[March 2026] Data File Update: The recommended workflow is now the standalone scripted pipeline under DPFunc_demo_pipeline/. Legacy long-form guidance (old tutorial/training sections) has been removed from this README.

[December 2025] Data File Update: As the old download links currently is unavailable, we have updated the dataset used in DPFunc paper here: https://drive.google.com/file/d/1qrxbkk450GJzhVfqnAN9Ms798owrq96n/view?usp=sharing​. It consists of protein id, sequences and corresponding GO terms. You can follow our tutorial to process AF-PDB, InterPro and ESM representations, so that you can retrain our model.

[June 2025] Data Processing Tutorial Update: We have streamlined our data processing pipeline with a comprehensive step-by-step tutorial. Users can now easily generate all required data using our new Jupyter notebook instead of following the previous complex workflow.

---

Dataset and Model Files

1. Benchmark Dataset (updated)

The dataset used in the DPFunc paper (protein IDs, sequences, GO terms):

• https://drive.google.com/file/d/1qrxbkk450GJzhVfqnAN9Ms798owrq96n/view?usp=sharing

2. Trained Model Weights

Pretrained DPFunc model weights:

• https://drive.google.com/file/d/1V0VTFTiB29ilbAIOZn0okBQWPlbOI3wN/view?usp=drive_link

Updated Usage (Recommended)

Use the standalone pipeline in:

• Chinese guide: ./DPFunc_demo_pipeline/README_zh.md
• English guide: ./DPFunc_demo_pipeline/README.md

Typical workflow:

1. Prepare a manifest (templates/manifest_template.tsv format).
2. Build a standalone workspace with build_data_demo.py.
3. Train or predict with run_dpfunc.py.

Example commands:

python DPFunc_demo_pipeline/scripts/build_data_demo.py \
  --manifest /path/to/your_manifest.tsv \
  --workspace /path/to/demo_workspace_mf \
  --ontology mf \
  --pdb-dir /path/to/pdb_folder \
  --inter-idx ./data/inter_idx.pkl \
  --go-obo /path/to/go.obo

python DPFunc_demo_pipeline/scripts/run_dpfunc.py train \
  --workspace /path/to/demo_workspace_mf \
  --ontology mf \
  --gpu 0 \
  --epoch 15 \
  --pre-name my_model

python DPFunc_demo_pipeline/scripts/run_dpfunc.py predict \
  --workspace /path/to/demo_workspace_mf \
  --ontology mf \
  --gpu 0 \
  --pre-name DPFunc

Notes

• Core model entry scripts remain unchanged: DPFunc_main.py, DPFunc_pred.py.

Contact

• Wenkang Wang: wangwk@csu.edu.cn
• Min Li: limin@mail.csu.edu.cn

Citation

Wang W, Shuai Y, Zeng M, et al. DPFunc: accurately predicting protein function via deep learning with domain-guided structure information. Nature Communications, 2025, 16(1): 70.