EnzEngDB

A comprehensive database and analysis pipeline for studying directed evolution of enzymes performing new-to-nature reactions.

Overview

The Enzyme Engineering database curates and analyzes data from directed evolution experiments documented in scientific literature, focusing on engineered enzymes that catalyze reactions not found in nature. The project creates molecular embeddings for both protein sequences and chemical reactions to enable machine learning applications and comparative analyses.

Features

• Data Curation: Systematic collection of enzyme-reaction pairs from 36+ research papers
• Molecular Embeddings: State-of-the-art embeddings for proteins (ESM3) and reactions (ChemBERTa2, RxnFP)
• Chemical Space Analysis: Visualization and comparison of engineered vs natural enzyme reaction space
• Standardized Format: Conversion to LevSeq format for broader accessibility
• Comprehensive Pipeline: End-to-end processing from raw data to analysis-ready datasets

Dataset Statistics

• 1,341 enzyme-reaction pairs
• 640 unique reactions
• 367 unique protein variants
• 36 research papers included

Installation

Prerequisites

• Python 3.8+
• PyTorch (for ESM models)
• RDKit (for chemistry operations)

Install via pip

# Clone the repository
git clone https://github.com/yourusername/EnzymeEngineeringDB.git
cd EnzymeEngineeringDB

# Create env
conda create --name enzengdb

# Install dependencies
pip install -r requirements.txt

Dependencies

The project requires the following main packages:

• Core: pandas, numpy, scikit-learn, matplotlib, seaborn
• Chemistry: rdkit, pubchempy, biopython
• Deep Learning: torch, esm, huggingface-hub
• Other: enzymetk, sciutil, sciviso

See requirements.txt for complete list with versions.

Usage

The analysis pipeline consists of four main notebooks that should be run in sequence:

1. Clean Reaction Data

jupyter notebook analysis/N1_CleanReactionData.ipynb

• Validates and canonicalizes reaction SMILES
• Creates reaction embeddings using ChemBERTa2 and RxnFP
• Outputs: cannoical_smiles.pkl, rxn_chemberta.pkl, rxn_rxnfp.pkl

2. Clean Enzyme Data

jupyter notebook analysis/N2_CleanEnzymeData.ipynb

• Processes enzyme sequences and mutations
• Generates protein embeddings using ESM3
• Outputs: protein-evolution-database_V4_embedded_proteins.pkl, variant_df_no_errors.pkl

3. Analyze Combined Data

jupyter notebook analysis/N3_AnalyseEnzymeReactionData.ipynb

• Combines protein and reaction data
• Performs PCA analysis and visualization
• Compares engineered enzymes to natural enzyme space

4. Convert to Standard Format

jupyter notebook analysis/N4_ConvertFormatToLevSeq.ipynb

• Converts data to LevSeq format
• Organizes by experiment/paper
• Creates metadata files

Project Structure

DirectedEvolutionDB/
---  README.md
---  requirements.txt
---  LICENSE
---  data/                    # Raw data files
---  nalysis/
------   N1_CleanReactionData.ipynb
------ N2_CleanEnzymeData.ipynb
------ N3_AnalyseEnzymeReactionData.ipynb
------ N4_ConvertFormatToLevSeq.ipynb
------  scripts/
------ esm3.py         # ESM3 embedding utilities
------ output/             # Processed data outputs
------ Archive/            # Previous notebook versions

Output Files

• cannoical_smiles.pkl: Standardized reaction SMILES
• rxn_chemberta.pkl: ChemBERTa2 reaction embeddings
• rxn_rxnfp.pkl: RxnFP reaction fingerprints
• protein-evolution-database_V4_embedded_proteins.pkl: ESM3 protein embeddings
• variant_df_no_errors.pkl: Cleaned variant data with yields

Key Findings

1. Directed evolution has successfully expanded enzyme function into previously unexplored chemical space
2. Engineered enzymes cluster in distinct regions when visualized using dimensionality reduction
3. Different research groups tend to explore different regions of chemical/sequence space
4. The database captures the diversity of new-to-nature enzymatic reactions

LLM pipeline

The LLM pipeline can be accessed at: https://github.com/YuemingLong/DEBase The Automated download of pubmed papers can be accessed at: https://github.com/31415erre/pubmed2pdf

Website and database

The database and website can be accessed at:

Contributing

We welcome contributions! Please feel free to submit issues or pull requests. For formatting please format your data in the LevSeq output format.

This means we require several headers.

Citation

To cite please refer to our releases.

A paper citation will be coming soon.

License

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

Contact

For questions or collaborations, please open an issue on GitHub.