lncrnapi 1.2

A CLI tool for predicting lncRNA–Protein interactions using transformer embeddings and CatBoost

🧬 lncrna-PI - LncRNA–Protein Interaction Prediction

lncrnaPI is a command-line tool for predicting lncRNA–Protein interactions using pre-trained language models (DNABERT-2 and ESM-2) for sequence embedding and a CatBoost classifier for interaction probability estimation.

It supports two modes:

• Rapid — based on sequence composition features (fast, lightweight)
• LLM — based on transformer embeddings (DNABERT2 + ESM2)

The script performs all-by-all predictions between every lncRNA and every protein sequence in the provided FASTA files.

---

📦 Features

• Vectorized and efficient FASTA parsing
• All-by-all pairing of lncRNA and protein sequences
• Automatic feature extraction:
  • Rapid mode: nucleotide and amino acid composition (%)
  • LLM mode: transformer-based embeddings (DNABERT2 + ESM2)
• Automatic model selection:
  • catboost_model_rapid.joblib → Composition model
  • catboost_dnabert2_esm-t30.joblib → Embedding model
• GPU-aware embedding generation (with safe fallback to CPU)
• Generates probability and binary interaction predictions

---

🧰 Dependencies

The tool was developed using Python 3.10. Install the following dependencies before running the script:

pip install torch==2.6.0 transformers==4.57.0 catboost==1.2.8 joblib tqdm numpy pandas

---

⚙️ Usage

1️⃣ Rapid (Composition-Based) Prediction

This mode uses simple % composition features (very fast).

python lncrnapi_cli.py   -lf ./data/example_lncRNA.fasta   -pf ./data/example_protein.fasta   -wd ./output   -model rapid

Model used:
./model/catboost_model_rapid.joblib

---

2️⃣ LLM (Embedding-Based) Prediction

This mode uses transformer embeddings from DNABERT2 (for lncRNA) and ESM2-T30 (for protein).

python lncrnapi_cli.py  -lf ./data/example_lncRNA.fasta   -pf ./data/example_protein.fasta   -wd ./output   -model llm

Model used:
./model/catboost_dnabert2_esm-t30.joblib

---

Arguments

Argument	Description	Required
-lf	Path to the FASTA file containing lncRNA sequences.	✅
-pf	Path to the FASTA file containing protein sequences.	✅
-wd	Path to the working directory.	✅
-model	Choice of model to be used.	✅
-t	Threshold	❌

---

💾 Output

A CSV file named output.csv is generated in the output directory:

lncRNA_ID	Protein_ID	Interaction_Probability	Predicted_Label
lnc1	P12345	0.87	1
lnc1	P67890	0.34	0
...	...	...	...

• Interaction_Probability: Probability predicted by CatBoost
• Predicted_Label: 1 → interaction, 0 → non-interaction

---

⚡ Hardware Acceleration

The script automatically detects and uses available hardware:

• ✅ CUDA GPU (NVIDIA)
• ✅ MPS (Apple Silicon)
• ⚠️ CPU (fallback)

---

🛠 Troubleshooting

Issue	Possible Cause	Solution
Model file not found	Wrong --model_path	Check the file path
No sequences found in FASTA	Invalid FASTA format	Ensure > headers are present
safetensors error	Missing library	Install with pip install safetensors
Slow performance	CPU usage	Use GPU-enabled environment

---

📜 Citation

If you use this tool in your research, please cite:

** Choudhury et al.**
Prediction of lncRNA-protein interacting pairs using LLM embeddings based on evolutionary information (2025)

---

🧩 Repository Structure

├── lncrnapi_cli.py       # Main CLI script
├── README.md
├── LICENSE
├── test_lncrna.fa
├── test_protein.fa
├── output.csv 
└── models/
    ├── catboost_model_rapid.joblib
    └── catboost_dnabert2_esm-t30.joblib                

---