AiZynthFinder

Overview

AiZynthFinder supports computer-assisted retrosynthetic route planning. Its default method uses Monte Carlo tree search to work backward from a target molecule toward purchasable precursors. A neural-network expansion policy proposes precursor transformations using a library of known reaction templates. It is therefore a planning toolkit that uses trained models, rather than a single standalone predictive model. The project also documents support for alternative search algorithms and expansion policies.

In a synthesis-planning workflow, the tool provides candidate routes for subsequent assessment. The documented command-line interface accepts a SMILES input file alongside a configuration file, while an application interface offers another way to access the algorithm. Running a search requires a stock file and a trained expansion policy network; a trained filter policy network is optional. The README links public assets and describes a download utility that also creates a configuration file for either interface.

The configured stock defines the precursor inventory available to the planning workflow, and the expansion policy supplies template-based suggestions during search. These inputs should be considered when interpreting proposed routes. The source excerpts do not establish laboratory feasibility, comparative planning accuracy, or performance on particular target classes. Environment support is also conditional on dependency availability: the README specifies Python 3.10–3.12 and describes Linux, Windows and macOS support with that qualification. The repository code has an MIT licence, but that evidence does not establish the terms for separately downloaded model or stock assets.

Key Features

• Default Monte Carlo tree search recursively decomposes target molecules toward purchasable precursors.
• Neural-network expansion policies propose precursors using known reaction templates.
• Supports multiple search algorithms and expansion policies rather than requiring only the default configuration.
• Provides `aizynthcli` and `aizynthapp` interfaces; the documented CLI workflow accepts configuration and SMILES files.
• Uses a required stock file and expansion policy network, with an optional trained filter policy network.
• Includes `download_public_data` to obtain public assets and generate a configuration file for the supplied interfaces.

Use Cases

• Suggested evaluation: explore candidate retrosynthetic routes for representative target molecules against a selected precursor stock, then assess the proposals separately.
• Suggested evaluation: compare documented search algorithms or expansion policies on the same target set to investigate their effect on route discovery.
• Suggested evaluation: assess whether the SMILES-file CLI workflow fits a batch synthesis-planning process before integrating it into a research pipeline.

How to Use

1. Start with the official documentation and repository README. Identify the intended interface and the configuration requirements before preparing targets.
2. Follow the README’s end-user installation procedure in a Python 3.10–3.12 environment. Choose the documented base package or broader installation according to the functionality needed; platform support depends on the dependencies.
3. Obtain a stock file and trained expansion policy network. The README links public planning assets and filter-policy assets. Check their separate terms; the filter network is optional.
4. Use the documented download_public_data utility if appropriate, or prepare the required assets and configuration following the documentation. The utility creates a config.yml file for either interface.
5. Prepare target SMILES and use aizynthcli, or open aizynthapp with the configuration. As an intended evaluation, inspect proposed routes against your chosen stock and assess chemical feasibility independently.