AiZynthFinder configuration

This page configures how AiZynthFinder searches for routes — target input, stock repositories, expansion and filter policies, search algorithm, and processing limits. Your stock and policy choices most affect the result: routes are only marked purchasable when all starting materials exist in the stocks you select, so start with uspto_50k and the stocks that reflect your actual procurement options.

Webapp input methods

The .bloom* webapp provides a streamlined interface for AiZynthFinder retrosynthesis analysis.

Target source selection

From previous job

• Job Output Selector: Browse completed jobs with compound outputs
• Compatible tools: REINVENT, ZINC, PubChem, other chemistry tools
• SMILES extraction: Automatic extraction from job outputs
• Search and pagination: Browse large result sets

Target SMILES input

• Required field: Target molecule in SMILES format
• Placeholder: "Enter target molecule SMILES string"
• Helper text: "SMILES representation of the target molecule for retrosynthesis"
• Validation: Form requires non-empty SMILES to submit

Basic options

Clustering toggle

• Enable clustering: Checkbox to group similar routes
• Purpose: Reduce redundancy in results
• Default: Disabled

Logging toggle

• Enable detailed logging: Checkbox for verbose output
• Purpose: Debug and analysis
• Default: Disabled

Advanced configuration (optional)

Toggle "Show advanced settings" to access additional parameters:

Configuration file

• Configuration File: Path to YAML configuration file
• Placeholder: config_local.yml
• Purpose: Override default search parameters

Processing options

Policy and filter configuration

Input molecule

SMILES input

• Required parameter: smiles - Target molecule in SMILES format
• Format: Single SMILES string or path to file containing SMILES
• Example: "CCc1ccc(cc1)C(C)C(=O)O" (ibuprofen)
• Validation: Must be valid, canonical SMILES string

Configuration file (optional)

• Parameter: config - Path to YAML configuration file
• Default: Uses config_local.yml if not specified
• Purpose: Override default search parameters with custom settings

Stocks and availability

Stock sources

• Parameter: stocks - Array of stock repository names
• Available options:
  • "zinc" - ZINC database (millions of compounds)
  • "emolecules" - eMolecules catalog
  • "custom" - User-uploaded building block libraries
  • "pubchem" - PubChem compounds
• Required: At least one stock source must be specified
• Impact: Routes are only marked "purchasable" if all starting materials exist in specified stocks

Stock selection strategy

• Start small: Begin with 1-2 relevant stock sources to keep search focused
• Match procurement: Choose stocks that reflect your actual purchasing options
• Combine sources: Use multiple stocks for broader coverage
• Update frequency: Refresh stock snapshots regularly as catalogs change

Policies and templates

Expansion policies

• Parameter: policy - Array of expansion policy names
• Available policies:
  • "uspto" - Default USPTO-trained policy
  • "uspto_50k" - Policy trained on 50k USPTO reactions (balanced speed/quality)
  • "uspto_500k" - Policy trained on 500k USPTO reactions (higher quality, slower)
• Default: Uses USPTO policy if not specified
• Recommendation: Start with uspto_50k for most use cases

Policy characteristics

• uspto: Fastest, good for initial screening
• uspto_50k: Better route quality, moderate speed
• uspto_500k: Highest quality routes, slowest search

Filters and post-processing

Route filters

• Parameter: filter - Array of filter names to apply
• Available filters:
  • "price" - Filter by reagent cost
  • "availability" - Filter by commercial availability
  • "toxicity" - Filter out toxic intermediates
  • "patent" - Check patent status
• Purpose: Remove undesirable routes from results

Filter policies (v4.0)

• Parameter: filter_policy - Machine learning-based reaction feasibility filters
• Purpose: Reject unlikely or infeasible reactions during search
• Benefit: Reduces search space and improves route quality by eliminating chemically implausible steps early
• Note: Filter policies were introduced in AiZynthFinder 4.0 (May 2024)

Bonds-to-freeze constraints (v4.0)

• Parameter: frozen_bonds - List of bond indices that should not be disconnected
• Format: Array of atom index pairs, e.g., [[0, 1], [5, 6]]
• Purpose: Preserve specific structural motifs during retrosynthesis
• Use cases:
  • Protect key pharmacophores from disconnection
  • Maintain stereochemistry at critical centers
  • Guide search toward preferred disconnection strategies
• Note: Bonds-to-freeze constraints were introduced in AiZynthFinder 4.0

Post-processing modules

• Parameter: post_processing - Array of post-processing module names
• Options: "cluster", "score", "validate"
• Default: Applies clustering and scoring by default

Pre-processing

• Parameter: pre_processing - Pre-processing module name
• Purpose: Prepare input molecule before search
• Default: Standard SMILES normalization

Search algorithm

Algorithm selection

• Parameter: search_algorithm - Search strategy for route exploration
• Available algorithms:
  • "mcts" - Monte Carlo Tree Search (default, best for most use cases)
  • "mcts_nested" - Nested Monte Carlo search (deeper exploration)
  • "greedy_best_first" - Greedy Best First Search (faster, less exhaustive)
• Default: Uses MCTS if not specified
• Recommendation: Start with mcts for balanced exploration; use greedy_best_first for quick screening of many molecules

Algorithm characteristics

• mcts: Balances exploration and exploitation, robust across molecule types
• mcts_nested: Better for complex molecules requiring deeper search trees
• greedy_best_first: Fastest option, prioritizes most promising paths immediately

Search parameters

Parallel processing

• Parameter: nproc - Number of processes to use
• Default: Uses available CPU cores
• Impact: More cores = faster search, higher memory usage

Clustering

• Parameter: cluster - Enable automatic route clustering
• Default: false
• Purpose: Group similar routes to reduce redundancy

Output options

• Parameter: output - Output file path for results
• Parameter: log_to_file - Enable detailed logging to file
• Default: false

Checkpointing

• Parameter: checkpoint - Path to checkpoint file
• Purpose: Resume interrupted searches
• Use case: Long-running searches that might be interrupted

Search limits and performance

Time management

• No explicit timeout parameter: Search runs until completion or resource limits
• Typical duration: 5-30 minutes depending on molecule complexity
• Scaling factors: Larger molecules, more stocks, complex policies = longer search

Result limits

• Route count: Limited to top-scoring routes (typically 50-200)
• Memory constraints: Large searches may hit memory limits before completion
• Quality over quantity: Focus on high-scoring routes rather than exhaustive search

Reproducibility and versioning

Configuration tracking

• Record parameters: Always document stock sources, policies, and filters used
• Version control: Model and policy versions are included in job metadata
• Consistent comparisons: Use identical configurations when comparing different targets

Metadata preservation

• Job artifacts: Include complete configuration in JSON metadata
• Audit trail: Policy versions and timestamps enable result validation
• Sharing: Share complete configuration for reproducible route generation

Recommended configurations

Quick screening

smiles: "CCc1ccc(cc1)C(C)C(=O)O"
stocks: ["zinc"]
policy: ["uspto_50k"]
nproc: 4

Comprehensive search

smiles: "CCc1ccc(cc1)C(C)C(=O)O"
stocks: ["zinc", "emolecules"]
policy: ["uspto_500k"]
filter: ["price", "availability"]
nproc: 8
cluster: true

Custom library search

smiles: "CCc1ccc(cc1)C(C)C(=O)O"
stocks: ["custom"]
policy: ["uspto_50k"]
filter: ["availability"]