antifp2 1.0.0

AntiFP2: A tool for prediction of Antifungal Proteins

AntiFP2

AntiFP2 is a tool for the prediction of antifungal proteins using a fine-tuned ESM2 language model, optionally enhanced by post-prediction adjustment with BLAST and MERCI motif detection.

This pipeline combines deep learning-based embeddings with classical bioinformatics methods for improved reliability in antifungal protein prediction.

---

🚀 Features

• Fine-tuned ESM2-t36_3B_UR50D model for antifungal prediction
• Post-prediction adjustment using:
  • BLAST: Sequence similarity matching to known antifungal/negative examples
  • MERCI: Motif Enrichment Recognition to enhance biological relevance
• One-by-one and batch prediction modes
• Rejection logging for low-quality or invalid sequences
• Hugging Face integration for model loading

---

📦 Installation

pip install git+https://github.com/patrik-ackerman/antifp2.git

Note: Requires Python ≥ 3.12

Ensure that BLAST+ and MERCI binaries are properly configured via the envfile as shown below.

---

📁 Project Structure

antifp2/
│
├── python_scripts/
│   ├── antifp2_ESM2.py      # Main pipeline with ESM2 + BLAST + MERCI
│   ├── antifp2_BLAST.py     # ESM2-only one-by-one predictor
│   └── envfile              # Config file for paths to BLAST and MERCI tools
│
├── MERCI/                   # MERCI motif files
├── blast_db/                # Preformatted BLAST database
├── README.md
├── setup.py
└── ...

---

🧪 Usage

🔮 ESM2-Only Prediction (one-by-one)

antifp2_blast --fasta path/to/input.fasta --output results.csv

• Output will include:

  • ID, probability, prediction columns

• Logs invalid sequences to rejected_log.txt

🧬 Full Pipeline (ESM2 + BLAST + MERCI)

antifp2_esm --fasta path/to/input.fasta --output ./output_dir/

• Performs predictions

• Runs BLAST against provided database

• Executes MERCI with motif file

• Adjusts predictions and saves final output to:

  • output_dir/<input>.adjusted.csv

Optional flag:

--no-cleanup   # Retains intermediate files like raw BLAST output, logs, etc.

---

🔧 Configuring Environment

The tool reads environment-specific paths from a file named envfile. Example format:

# Path settings for different OS
BLAST_ubuntu=/usr/bin/blastp
BLAST_windows=C:/Program Files/NCBI/blastp.exe
BLAST_macos=/usr/local/bin/blastp

BLAST_database=antifp2/blast_db/antifungal_db
MERCI=antifp2/MERCI/merci
MERCI_motif_file=antifp2/MERCI/motifs.motif

Make sure this file is located in antifp2/python_scripts/envfile.

---

📋 Output Format

adjusted.csv columns:

Column	Description
ID	Sequence ID from FASTA
probability	Raw ESM2-based antifungal probability
blast_adjustment	Adjustment based on BLAST hit
motif_adjustment	Adjustment based on MERCI hit
combined	Final adjusted probability
prediction	1 if combined ≥ 0.5, else 0

---

💾 Model Files

Downloaded automatically from Hugging Face:

• config.json
• pytorch_model.bin
• alphabet.bin

Repo: raghavagps-group/antifp2

---

📝 License

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

---

👨‍🔬 Author

Pratik Shinde Indian Institute of Information Technology Delhi Email

---

🌐 Links

• 🔗 GitHub: https://github.com/patrik-ackerman/antifp2
• 🤗 Hugging Face Model: raghavagps-group/antifp2