AiZynthFinder

AiZynthFinder is a retrosynthetic planning tool. Its default algorithm uses a Monte Carlo tree search that recursively breaks down a molecule into purchasable precursors.

The search is guided by a policy that suggests possible precursors using a neural network trained on known reaction templates. The tool is highly customizable, supporting various search algorithms and expansion policies.

The behavior of aizynthcli changes based on whether you provide one or multiple SMILES:

Single SMILES:

• Statistics are printed directly to the terminal.
• Top-ranked routes are saved to a JSON file (default: trees.json).
• The stock_info and trees columns are NOT included in the terminal output.

Multiple SMILES (Batch):

• Results are saved to a JSON or HDF5 file (default: output.json.gz).
• The output file contains all columns defined in the batch output schema (including stock_info and trees).

The promising_child() method of the MctsNode class is responsible for navigating the tree based on the Upper Confidence Bound (UCB) score.

• Selection Logic: It sorts children by their UCB score and selects the highest-scoring child for instantiation.
• Instantiation: When a child is selected, its associated RetroReaction is applied to create new precursors, which form the state of the new MctsNode.
• Failure Handling: If a reaction fails to produce precursors, or if a filter policy rejects the reaction, the child's value is set to a large negative value to prevent future selection.
• Recursion: The method calls itself recursively until a valid child can be instantiated. If no children can be instantiated, the is_expanded and is_expandable flags are updated, and the method returns None.

To run retrosynthesis experiments, you need a trained policy model and a stock collection. You can download publicly available data (a USPTO-based model and a ZINC database stock collection) to your current directory using the provided utility command.

This data allows you to use the config.yml file immediately with the package interfaces.

download_public_data

You can perform tree searches on single compounds using a graphical user interface within a Jupyter notebook.

To start, launch Jupyter:

jupyter notebook

In a new or existing notebook, instantiate the AiZynthApp class by providing the path to your configuration file. Executing this cell will launch the GUI.

Steps to run a search:

1. Execute the initialization cell (press Ctrl+Enter).
2. Enter the target SMILES string.
3. Select your desired stocks and policy model.
4. Click Run Search to begin the tree search.
5. Click Show Reactions to view the top-ranked routes.
6. You can optionally select and sort top-ranked routes using alternative scoring functions.

from aizynthfinder.interfaces import AiZynthApp
app = AiZynthApp("/path/to/configfile.yaml")

To perform a basic tree search, create a config.yml file that maps your expansion models and stock files to specific keys.

For a simple setup, you need:

1. Expansion: A list containing a Keras expansion model (ONNX format) and a template file (CSV.GZ or HDF5).
2. Stock: A path to an HDF5 stock file.

Example config.yml for a basic setup:

expansion:
  full:
    - uspto_expansion.onnx
    - uspto_templates.csv.gz
stock:
  zinc: zinc_stock.hdf5

To install AiZynthFinder, use Conda to create a Python 3.10-3.12 environment and then install the package via pip. You can choose between a full installation with all functionalities or a smaller package.

Full installation (recommended):

conda create "python>=3.10,<3.13" -n aizynth-env
conda activate aizynth-env
python -m pip install aizynthfinder[all]

Minimal installation:

conda create "python>=3.10,<3.13" -n aizynth-env
conda activate aizynth-env
python -m pip install aizynthfinder

The Chemformer expansion model uses a REST API to perform expansions.

Prerequisites:

1. Install the chemformer package.
2. Launch the chemformer REST API service.

Configuration: Add the following to your configuration file. Ensure the url points to the correct host and port where your REST service is running (defaulting to http://localhost:8000/chemformer-api/predict).

It is recommended to set a time_limit (e.g., 300 seconds) to allow the more computationally expensive expansion model enough time to complete sufficient retrosynthesis iterations.

expansion:
    chemformer:
        type: expansion_strategies.ChemformerBasedExpansionStrategy
        url: http://localhost:8000/chemformer-api/predict
search:
    algorithm_config:
        immediate_instantiation: [chemformer]
    time_limit: 300

To use a custom stock with the aizynthcli tool:

1. Save your custom class in a module (e.g., custom_stock.py).
2. Ensure the module is in a directory known to the Python interpreter.
3. Define an instance of your class named stock within that module.

Example custom_stock.py:

from rdkit.Chem import Lipinski
from aizynthfinder.context.stock.queries import StockQueryMixin

class CriteriaStock(StockQueryMixin):
    def __contains__(self, mol):
        return Lipinski.HeavyAtomCount(mol.rd_mol) < 10

stock = CriteriaStock()