Fungtion

Predict fungal effectors from FASTA files using ESM-1b embeddings and R SVM models.

Paper: https://doi.org/10.1016/j.jmb.2024.168613

For sequences predicted as fungal effectors, this project also generates:

• ESM-1b-based similarity network visualization
• relationship tree visualization
• an HTML report in the style of the original Fungtion web page

Clone the Repository

git clone https://github.com/noHup-cc/fungtion.git
cd fungtion

Install

Option 1: create the full Conda environment from environment.yml (recommended)

conda env create -f environment.yml
conda activate fungtion

# optional: install Fungtion as a local editable package
pip install -e .

Option 2: create the Conda environment manually

conda create -n fungtion -c conda-forge python=3.10 r-base=4.4 r-e1071 r-caret r-optparse
conda activate fungtion

# install the Python package in editable mode
pip install -e .

Developer tooling:

pip install -e ".[dev]"
pre-commit install
pre-commit run --all-files

PyTorch Compatibility

The default installation will install torch automatically.

If your system has specific CPU/GPU or CUDA driver requirements, you may need to reinstall a compatible PyTorch build manually after installation. For example:

# CPU only
pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cpu

# Example: CUDA 12.6
pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cu126

Please choose the PyTorch build that matches your local hardware and driver setup:

• https://pytorch.org/get-started/locally/

Example Run

Example FASTA:

• data/examples/example.fasta

Download ESM-1b Weights

Before prediction, you can download the pretrained ESM-1b weights through a separate setup step. This downloads the same ESM-1b model file that Fungtion would otherwise fetch automatically through fair-esm:

fungtion setup-models

By default, the downloaded file is stored at:

~/.cache/fungtion/models/esm1b_t33_650M_UR50S.pt

You can also choose a custom directory:

fungtion setup-models --model-dir /path/to/models

If the weights already exist locally, the setup step will skip downloading by default. To force a fresh download and overwrite the existing local copy, use:

fungtion setup-models --force

Simple:

fungtion \
  --fasta data/examples/example.fasta \
  --output-dir outputs \
  --prefix example_prediction \
  --device auto \
  --skip-visualization

With HTML report:

fungtion \
  --fasta data/examples/example.fasta \
  --output-dir outputs \
  --prefix example_prediction \
  --device auto \
  --html-report

Keep intermediate files:

fungtion \
  --fasta data/examples/example.fasta \
  --output-dir outputs \
  --prefix example_prediction \
  --device auto \
  --skip-visualization \
  --keep-temp

With GPU:

fungtion \
  --fasta data/examples/example.fasta \
  --output-dir outputs \
  --prefix example_prediction \
  --device cuda \
  --skip-visualization

Note

If the setup-downloaded weights exist in the default Fungtion cache directory, the prediction command will use them automatically.

If you pass --pretrain, Fungtion will use that explicit local path in preference to the default setup location.

You can also download the model file manually to a local path and run Fungtion with --pretrain /path/to/esm1b_t33_650M_UR50S.pt.

fungtion \
  --fasta data/examples/example.fasta \
  --output-dir outputs \
  --prefix example_prediction \
  --pretrain /path/to/esm1b_t33_650M_UR50S.pt \
  --device auto \
  --skip-visualization

Official ESM-1b weights:

• https://dl.fbaipublicfiles.com/fair-esm/models/esm1b_t33_650M_UR50S.pt

Example Run Output

The HTML example above generates everything under outputs/example_prediction/:

• example_prediction.csv: prediction results
• example_prediction_analysis/: network and tree intermediate outputs
• example_prediction.html: HTML report
• example_prediction_assets/: HTML report assets and per-sequence visualization pages
• example_prediction_temp_folder/: intermediate feature files when --keep-temp is used

Example Run Output CSV Columns

• header: FASTA header
• score: prediction score
• decision: yes or no
• type: Exp. or Pred.
• type_link: external link for Exp. entries when available

Notes

• If an input sequence is identical to a reference positive sequence, it is marked as Exp. and its score is shown as 1.
• Other sequences are marked as Pred. and scored by the R SVM models.
• The HTML report supports search, sorting, pagination, column toggling, export, help popovers, and visualization links.
• The ESM-1b network page supports clicking non-query nodes to open UniProt or NCBI Protein pages.
• Before committing code changes, run pre-commit run --all-files to apply Ruff fixes/formatting and basic whitespace checks.

More Parameters On Run

• --fasta: input FASTA file
• --output-dir: directory where the prefixed output folder will be created
• --prefix: naming identifier for the output folder and generated files
• --pretrain: optional local path to pretrained ESM-1b weights
• --device: device for ESM-1b feature extraction; choose from auto, cpu, or cuda
• --html-report: generate the HTML report and bundled assets
• --skip-visualization: skip network and tree generation
• --keep-temp: keep intermediate temporary files under <output-dir>/<prefix>/<prefix>_temp_folder/

Data

data/

• examples/: example FASTA files for demonstration and example runs

Full training and independent benchmark datasets used in the paper are available on Zenodo:

• https://zenodo.org/records/20808434
• DOI: 10.5281/zenodo.20808434

src/fungtion/reference_data/

• FungalEffector_positive.fasta: experimentally validated fungal effector proteins
• FungalEffector_positive_esm_uniref50.csv: ESM-1b features of the experimentally validated fungal effector proteins
• FungalEffector_positive_similarity_matrix/: precomputed reference similarity matrices used for network and relationship tree visualization

Citation

If you use this project, please cite:

@article{li2024fungtion,
  title={Fungtion: A Server for Predicting and Visualizing Fungal Effector Proteins},
  author={Li, Jiahui and Ren, Jinzheng and Dai, Wei and Stubenrauch, Christopher and Finn, Robert D. and Wang, Jiawei},
  journal={Journal of Molecular Biology},
  volume={436},
  number={17},
  pages={168613},
  year={2024},
  doi={10.1016/j.jmb.2024.168613}
}