mol-healer 0.1.5

Hit Expansion to Advanced Leads Using Enumerated Reactions

HEALER

Hit Expansion to Advanced Leads Using Enumerated Reactions

Installation • Building Blocks • Quick Start • Usage • Web Interface • Citation

---

HEALER generates synthetically accessible molecular analogs by combining retrosynthetic fragmentation with commercially available building blocks and validated reaction templates. It bridges the gap between computational design and laboratory synthesis.

Features

• Molecule HEALER — Retrosynthetically fragment a molecule and re-enumerate with similar building blocks
• Fragment HEALER — Enumerate from pre-fragmented molecules (multi-component SMILES)
• Site HEALER — Targeted enumeration at specific reactive sites with property filters
• Synthetically Accessible — All products use validated reaction templates
• Flexible — Works with any building block library in SDF format

Installation

Prerequisites

• Python ≥ 3.11
• Conda (recommended)

Option 1: pip install (recommended)

pip install mol-healer

Option 2: From source

git clone https://github.com/eneskelestemur/healer.git
cd healer

# Create conda environment
conda env create -f environment.yml
conda activate healer

# Install in development mode
pip install -e .

Optional: Web interface

pip install mol-healer[web]

Building Blocks

HEALER includes a small test set for demos. For production use, you'll need to set up building block libraries.

Download & Process

1. Download building blocks (e.g., from Enamine)

2. Preprocess to add reaction annotations:

   preprocess-bb ~/Downloads/Enamine_BBs.zip -o ~/.healer/buildingblocks/ --verbose
   

3. Set the data directory:

   export HEALER_DATA_DIR=~/.healer
   

Custom Libraries

Any SDF file can be used as a building block source. Just preprocess it:

preprocess-bb my_custom_library.sdf -o ~/.healer/buildingblocks/

Then reference it by path:

healer = MoleculeHEALER(bb_source='/path/to/my_custom_library_processed.sdf')

Quick Start

Python API

from healer import MoleculeHEALER

# Initialize with building block source and reaction filters
healer = MoleculeHEALER(
    bb_source='test',  # Use test set or path to processed SDF
    reaction_tags=['amide coupling', 'N-arylation'],
    sim_threshold=0.5,
)

# Set query molecule and enumerate
healer.set_query_mol("CC1(C)SC2C(NC(=O)Cc3ccccc3)C(=O)N2C1C(=O)O", n_compositions=10)
healer.enumerate(max_evals_per_comp=500)

# Get results
results = healer.get_results(calc_similarity=True, calc_properties=True)
print(f"Generated {len(results)} analogs")

Command Line

# Basic enumeration
healer molecule "CCO" --bb-source test -o results.csv

# View molecule with atom indices (for site specification)
healer view "c1ccccc1N"

# Site-specific enumeration
healer site "c1ccccc1N" --reactive-sites "[5]" --bb-source test

# Fragment-based enumeration
healer fragment "c1ccccc1.CC(=O)O" --bb-source test

# Parallel synthesis loop (uses all CPUs)
healer molecule input.csv --n-jobs -1 -o results.csv

Contributing Reactions

We welcome contributions to expand our reaction library! Each reaction entry in reactions.json follows this format:

{
  "reaction-name": {
    "description": "Brief description of the reaction mechanism",
    "long_name": "Full reaction name",
    "syn_smarts": "[reactant1].[reactant2]>>[product]",
    "retro_smarts": "[product]>>[fragment1].[fragment2]",
    "rhs_classes": ["bb-class-1", "bb-class-2"],
    "tags": ["reaction-type", "functional-group"],
    "tier": 1
  }
}

Key fields: syn_smarts (forward reaction), retro_smarts (retrosynthetic transform), rhs_classes (building block functional group classes), and tags (for filtering).

To contribute a reaction, please open a GitHub issue with your proposed reaction SMARTS, or contact [e-mail].

Usage

HEALER Classes

Class	Use Case	Input
MoleculeHEALER	Full retrosynthetic enumeration	Single molecule SMILES
FragmentHEALER	Enumeration from fragments	Dot-separated SMILES
SiteHEALER	Site-specific enumeration	Molecule + atom indices

Key Parameters

Parameter	Description	Default
Common (all modes)
bb_source	Building block library (path to processed SDF)	'test'
reaction_tags	Filter reactions by tag (list or 'all')	'all'
shuffle_bb_order	Randomize building block order	False
max_evals_per_comp	Max reaction attempts per composition	None (unlimited)
max_products_per_comp	Max products per composition	None (unlimited)
max_total_products	Stop after this many total products	None (unlimited)
MoleculeHEALER / FragmentHEALER
sim_threshold	Minimum Tanimoto similarity for BB matching	0.5
max_bbs_per_frag	Max BBs per fragment (-1 = use threshold)	-1
n_compositions	Number of fragment compositions to explore	10
retro_tree_depth	Depth of retrosynthetic tree search	1
min_frag_size	Minimum fragment size in heavy atoms	3
randomize_compositions	Shuffle composition order	False
random_seed	Seed for reproducibility (-1 = random)	-1
custom_split_sites	Manual bond indices to break (skips retro)	None
SiteHEALER
reactive_sites	Atom indices for enumeration (list of ints)	None (all sites)
rules	Property filters, e.g., {'MW': (0, 500)}	{}
struct_rules	Required SMARTS patterns in BBs	[]

CLI Commands

healer molecule <input> [options]   # Molecule-based enumeration
healer site <input> [options]       # Site-specific enumeration  
healer fragment <input> [options]   # Fragment-based enumeration
healer view <smiles>                # Visualize with atom indices

Run healer <command> --help for detailed options.

Web Interface

HEALER includes a web UI for interactive use.

Local Mode (Simple)

# Start the server (no Redis needed)
healer-ui

Open http://localhost:8000 in your browser.

Server Mode (Production)

For deployments with multiple users, use Celery/Redis for async job processing:

# Terminal 1: Redis
redis-server

# Terminal 2: Celery worker
celery -A healer.web.celery_worker worker --loglevel=info

# Terminal 3: Backend
HEALER_SERVER_MODE=true healer-ui

See web_client/README.md for development setup.

Configuration

Copy .env.example to .env and customize:

cp .env.example .env

Key settings:

• HEALER_SERVER_MODE — Enable async job processing
• HEALER_DATA_DIR — Custom data directory
• HEALER_LIMIT_* — Server parameter limits

Project Structure

healer/
├── healer/                 # Core package
│   ├── application/        # HEALER classes
│   ├── domain/             # Data models
│   ├── utils/              # Utilities
│   ├── web/                # FastAPI backend
│   ├── scripts/            # CLI scripts
│   └── data/               # Bundled data (reactions, test BBs)
├── web_client/             # React frontend
├── tests/                  # Test suite
└── benchmark/              # Benchmarking scripts

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Development Setup

git clone https://github.com/eneskelestemur/healer.git
cd healer
conda env create -f environment.yml
conda activate healer
pip install -e ".[web]"

# Run tests
pytest tests/

# Start frontend dev server
cd web_client && npm install && npm run dev

Citation

If you use HEALER in your research, please cite:

@article{healer2025,
  title={HEALER: Hit Expansion to Advanced Leads Using Enumerated Reactions},
  author={Kelestemur, Enes and ...},
  journal={...},
  year={2025}
}

License

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

Acknowledgments

• Reaction template formats mainly adapted from datamol
• Building block preprocessing inspired by retrosynthesis literature
• Ketcher for molecular drawing

---

Report Bug • Request Feature