jackkuo/BCE-Vir-Prediction_model

BCE-Vir-Prediction

A virus epitope prediction tool based on ESM (Evolutionary Scale Modeling). This tool uses a pre-trained ESM classification model to perform sliding window predictions on protein sequences, identifying potential antigen epitopes and functional domains.

Features

• Epitope Prediction (bcepre_predict_logits.py): Uses a pre-trained ESM classification model to split protein sequences with sliding windows, performs classification predictions on each subsequence (e.g., whether it is an antigen epitope, functional domain, etc.), and saves prediction results along with corresponding logits values.
• Amino Acid Probability Prediction (bcepre_predict_softmax.py): Converts sliding window prediction results into probability values aggregated by amino acid position, outputting a results table containing amino acid types, epitope probabilities, and coverage counts.

Model

The pre-trained model can be downloaded from Hugging Face:

• Model Repository: jackkuo/BCE-Vir-Prediction_model

• Code Repository: JackKuo666/BCE-Vir-Prediction

Model Download Instructions

This folder is used to store the trained ESM model files.

How to Download the Model

Method 1: Using Hugging Face Hub (Recommended)

Use the huggingface_hub library to download the model:

pip install huggingface_hub

Then run the following Python code:

from huggingface_hub import snapshot_download

# Download the model to the current folder
snapshot_download(
    repo_id="jackkuo/BCE-Vir-Prediction_model",
    local_dir="./",
    local_dir_use_symlinks=False
)

Or use huggingface-cli in the command line:

huggingface-cli download jackkuo/BCE-Vir-Prediction_model --local-dir ./ --local-dir-use-symlinks False

Method 2: Using Git LFS

If Git LFS is installed, you can clone directly:

git lfs install
git clone https://huggingface.co/jackkuo/BCE-Vir-Prediction_model .

Method 3: Manual Download

Visit the model page: https://huggingface.co/jackkuo/BCE-Vir-Prediction_model

Select the required files from the file list to download and save them to this folder.

Model File Structure

After downloading, this folder should contain the following files:

• config.json - Model configuration file
• model.safetensors - Model weights file (in safetensors format)
• tokenizer_config.json - Tokenizer configuration file
• vocab.txt - Vocabulary file
• special_tokens_map.json - Special tokens mapping file

Usage

Step 1: Download the Model

First, download the pre-trained model to the trained_esm_model folder.

Step 2: Prepare Input Files

Place the protein sequence file (FASTA format) to be predicted in the example_data folder, or modify the input file path in the script.

Step 3: Run Epitope Prediction

Run the bcepre_predict_logits.py script for epitope prediction:

python bcepre_predict_logits.py

This script will:

• Read the protein sequence file in FASTA format
• Split the sequence using sliding windows (default minimum window size is 5)
• Perform classification predictions on each subsequence
• Output a CSV file containing the following fields:
  • sequence: Subsequence
  • window_size: Window size
  • prediction: Predicted class
  • logit_0, logit_1, ...: Logits values for each class

Output files are saved in the predictions/ folder by default.

Step 4: Calculate Amino Acid Position Probabilities

Run the bcepre_predict_softmax.py script to convert prediction results into aggregated probabilities by amino acid position:

python bcepre_predict_softmax.py

This script will:

• Read the CSV file generated by bcepre_predict_logits.py
• Calculate epitope probability for each subsequence (using softmax function)
• Aggregate probability values by amino acid position
• Output a CSV file containing the following fields:
  • position: Amino acid position (starting from 1)
  • amino_acid: Amino acid type
  • probability: Epitope probability at this position (average of all window predictions covering this position)
  • coverage: Number of windows covering this position

License

This project is licensed under the MIT License. See the LICENSE file for details.

Citation

If you use this tool for research, please cite the relevant models and code repositories.

Contact

For questions or suggestions, please contact us through GitHub Issues.