rxnrecer 1.2.0

Deep learning framework for predicting enzyme-catalyzed reactions from protein sequences

RXNRECer

RXNRECer v1.2.0 is a deep learning framework for predicting enzyme-catalyzed reactions from protein sequences. It is the official implementation of "RXNRECer: Active Learning with Protein Language Models for Fine-Grained Enzyme Reaction Prediction."

🚀 Features

• Multi-Stage Prediction: S1 (reaction prediction), S2 (reaction integration), S3 (LLM reasoning)
• Protein Sequence Analysis: Process protein sequences in FASTA format
• Deep Learning Models: ESM-2 embeddings with advanced neural architectures
• GPU Acceleration: CUDA support for faster inference
• Easy-to-use CLI: Simple command-line interface with comprehensive options

📋 Requirements

• Python 3.10+
• PyTorch 2.0+
• CUDA 11.0+ (recommended)
• 32GB+ RAM
• 40GB+ disk space

🚀 Quick Start

1. Install

# Install from PyPI (recommended)
pip install rxnrecer

# Or install from GitHub
pip install git+https://github.com/kingstdio/RXNRECer.git

2. Download Data

# Download required data and model files (~20.5GB total)
rxnrecer-download-data

# Or download separately
rxnrecer-download-data --data-only      # ~8.6GB
rxnrecer-download-data --models-only    # ~11.9GB

3. Run Prediction

# Basic prediction (S1 mode)
rxnrecer -i input.fasta -o output.tsv -m s1

# Detailed prediction (S2 mode)
rxnrecer -i input.fasta -o output.tsv -m s2

# LLM reasoning (S3 mode, requires API key)
rxnrecer -i input.fasta -o output.json -m s3 -f json

🔧 Usage

Command Line Options

rxnrecer [OPTIONS]

Options:
  -i, --input_fasta    Input FASTA file path (required)
  -o, --output_file    Output file path
  -f, --format         Output format: tsv or json (default: tsv)
  -m, --mode           Prediction mode: s1, s2, or s3 (default: s1)
  -b, --batch_size     Batch size for processing (default: 100)
  -c, --cache          Enable caching (default: enabled)
  -v, --version        Show version

Examples

# Basic usage
rxnrecer -i proteins.fasta -o results.tsv

# Custom batch size
rxnrecer -i proteins.fasta -o results.tsv -b 50

# JSON output
rxnrecer -i proteins.fasta -o results.json -f json

# Disable cache (by default, caching is enabled)
rxnrecer -i proteins.fasta -o results.tsv

Input Format

FASTA file with protein sequences:

>P12345|Sample protein 1
MKTVRQERLKSIVRILERSKEPVSGAQLAEELSVSRQVIVQDIAYLRSLGYNIVATPRGYVLAGG
>P67890|Sample protein 2
MKLIVWALLLLAAWAVERSKEPVSGAQLAEELSVSRQVIVQDIAYLRSLGYNIVATPRGYVLAGG

Output Formats

TSV Output (S1/S2):

input_id	RXNRECer	RXNRECer_with_prob	rxn_details
P12345	RHEA:24076;RHEA:14709	0.9999;0.9999	[reaction details]

JSON Output (S3):

[
  {
    "reaction_id": "RHEA:24076",
    "prediction_confidence": 0.9999,
    "reaction_details": {...}
  }
]

🆕 Features

Smart Caching

Results are automatically cached for faster repeated predictions:

# Check cache status
rxnrecer-cache status

# Clear cache
rxnrecer-cache clear --all

Data Management

Easy data and model file management:

# Download data
rxnrecer-download-data

# Force re-download
rxnrecer-download-data --force

📁 Project Structure

rxnrecer/                    # Main Python package
├── cli/                     # Command-line interface
├── config/                  # Configuration
├── lib/                     # Core libraries
│   ├── datasource/          # Data source handling
│   ├── embedding/           # Protein embeddings
│   ├── llm/                 # Language model integration
│   ├── ml/                  # Machine learning utilities
│   ├── model/               # Model architectures
│   ├── rxn/                 # Reaction processing
│   └── smi/                 # SMILES handling
├── models/                  # Neural network models
└── utils/                   # Utility functions

data/                        # Data files (download required)
├── chebi/                   # ChEBI database
├── cpd_svg/                 # Compound SVG files
├── datasets/                # Training datasets
├── dict/                    # Dictionary files
├── feature_bank/            # Feature bank
├── rhea/                    # RHEA database
├── rxn_json/                # Reaction JSON files
├── sample/                  # Sample data
└── uniprot/                 # UniProt database

ckpt/                        # Model checkpoints (download required)
├── prostt5/                 # ProSTT5 model files
└── rxnrecer/                # RXNRECer model files

results/                     # Output results
├── cache/                   # Prediction cache
├── logs/                    # Log files
├── predictions/             # Prediction outputs
└── sample/                  # Sample results

docs/                        # Documentation
scripts/                     # Build and utility scripts

🔧 Configuration

For S3 mode (LLM reasoning), set your API key:

export LLM_API_KEY="your_api_key_here"
export LLM_API_URL="https://openrouter.ai/api/v1"

📚 Documentation

• Installation Guide - Detailed setup instructions
• Release Notes - Version information

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Open a Pull Request

📄 License

MIT License - see LICENSE file for details.

📞 Contact

• Author: Zhenkun Shi
• Email: [e-mail]
• Project: https://github.com/kingstdio/RXNRECer
• PyPI: https://pypi.org/project/rxnrecer/