path-finder-retrosynthesis 1.0.3

Retrosynthesis route finder — AiZynthFinder + Rxn-INSIGHT + Chemistry by Design

Path Finder

Retrosynthesis route finder — AiZynthFinder · Rxn-INSIGHT · Chemistry by Design

Yara Chahda · Corentin Postmann · Inès Ouchen — EPFL 2026

---

User installation

1. Install RDKit

RDKit cannot be installed via pip — conda is required for this one step.

conda install -c conda-forge rdkit

2. Install Path Finder

pip install path-finder-retrosynthesis

3. Run the setup wizard

path-finder-setup

This automatically:

• copies the bundled datasets into data/
• downloads the AiZynthFinder model files (~500 MB) via the official AiZynthFinder downloader
• generates data/config.yml with the correct paths

If the automatic download fails, download the model files manually from https://github.com/MolecularAI/aizynthfinder/releases and place them in data/aizynthfinder/.

4. Download the Rxn-INSIGHT USPTO database

Download uspto_rxn_insight.gzip from: The rxn-INSIGHT article

Place it in data/uspto_rxn_insight.gzip.

This file enables reaction condition prediction for novel routes (predicted routes section). Without it, only dataset and validated routes are shown.

5. Launch

path-finder

Open http://localhost:8501 in your browser.

---

Summary

conda install -c conda-forge rdkit
pip install path-finder-retrosynthesis
path-finder-setup
# → place uspto_rxn_insight.gzip in data/
path-finder

---

What the app does

Path Finder finds and ranks retrosynthesis routes for a target molecule using three sources:

Section	Source	Conditions	Yield in scoring
📚 Dataset	Curated Chemistry by Design routes	Real	Yes
✅ Validated	AiZynthFinder + generic reactions (USPTO)	Real	Yes
🤖 Predicted	AiZynthFinder + Rxn-INSIGHT	Predicted	No

Routes are scored using a weighted 1/i² scheme across three user-chosen criteria: steps, yield, atom economy, E-factor, or safety.

---

Data files

File	Bundled	Description
reaction_dataset.json	✅	Curated synthesis routes
toxicity_dataset.json	✅	Safety scores for reagents and solvents
generic_reactions.json	✅	10 000 USPTO reactions for step validation
data/aizynthfinder/	❌	AiZynthFinder model files — downloaded by wizard
data/config.yml	❌	Generated by wizard — do not commit
data/uspto_rxn_insight.gzip	❌	Rxn-INSIGHT USPTO database — download manually

---

Troubleshooting

Problem	Solution
config.yml not found	Run path-finder-setup
AiZynthFinder crash	Check that all paths in data/config.yml are absolute
No routes found	Try Galanthamine (OC1C=C[C@@]23c4cc(OC)ccc4CN(C)C[C@@H]2[C@@H]1O3)
Predicted routes disabled	Add data/uspto_rxn_insight.gzip (see step 4 above)
Slow search (~2 min)	Normal — AiZynthFinder MCTS is computationally intensive

---

Developer setup

git clone https://github.com/YaraChahda/path_finder.git
cd path_finder
conda install -c conda-forge rdkit
pip install -e .
path-finder-setup
path-finder

Project structure

path_finder/
├── src/path_finder/
│   ├── app_path_finder.py      # Streamlit front-end
│   ├── route_engine.py         # Scoring, AiZynthFinder, Rxn-INSIGHT
│   ├── molecule_rendering.py   # RDKit Cairo rendering
│   ├── localization.py         # EN/FR UI strings
│   ├── report_builder.py       # PDF generation
│   ├── cli.py                  # path-finder and path-finder-setup commands
│   ├── assets/banner.png
│   └── data/                   # bundled datasets + config template
├── data/                       # working data directory (not committed)
│   ├── config.yml              # generated by path-finder-setup
│   ├── aizynthfinder/          # model files downloaded by path-finder-setup
│   └── uspto_rxn_insight.gzip  # download manually
├── tests/
├── pyproject.toml
└── README.md

Publishing a new version

sed -i '' 's/version = "X.Y.Z"/version = "X.Y.Z+1"/' pyproject.toml
git add pyproject.toml
git commit -m "release: vX.Y.Z+1"
git tag vX.Y.Z+1
git push origin main --tags
# GitHub Actions publishes to PyPI automatically

Running tests

pytest tests/

---

Citation

• AiZynthFinder: Genheden et al., J. Cheminf. 2020 — doi:10.1186/s13321-020-00472-1
• Rxn-INSIGHT: Thakkar et al., J. Cheminf. 2023 — doi:10.1186/s13321-023-00744-4
• Open Reaction Database: Kearnes et al., JACS 2021 — doi:10.1021/jacs.1c09820