SynthSense: Synthesis-aware REINVENT Rewards

November 25, 2025 · View on GitHub

SynthSense Logo

SynthSense is a scoring component for REINVENT4 that evaluates molecules based on their synthetic feasibility and route coherence using retrosynthetic analysis from AiZynthFinder.

For details please check the preprint.

Features

SynthSense provides multiple rewards to guide molecular generation:

  • SFScore: Synthetic feasibility score combining stock availability, reaction preference and route complexity
  • RRScore: Reference route similarity score using Tree Edit Distance (TED)
  • Route Popularity: Promotes coherent synthetic routes within batches
  • Fill-a-Plate: Promotes coherent synthetic routes across batches (route diversity filter)

Prerequisites

SynthSense requires the following software:

  1. REINVENT4 - Molecular generation framework
    • Important: Required version 4.7 or newer, with scoring cache disable option (required for SynthSense)
  2. AiZynthFinder - Retrosynthetic planning tool
  3. NameRxn (NextMove Software) - Reaction classification tool
    • Commercial software - purchase required from NextMove Software
    • Required for annotating reaction classes in AiZynthFinder
    • See NAMERXN_SETUP.md for detailed setup instructions

Note: NameRxn is critical for SynthSense functionality as it provides reaction class annotations.

Installation

1. Install Prerequisites

Ensure REINVENT4 (4.7 or newer) and AiZynthFinder are installed and working in your environment.

Critical: SynthSense sets no_cache = True to disable scoring cache, which is essential for proper functionality. Ensure your REINVENT4 installation supports this feature.

2. Set up NameRxn

Follow the instructions in NAMERXN_SETUP.md to configure NameRxn for annotating AiZynthFinder retrosynthetic trees.

3. Install SynthSense

Synthsense (initial release) already comes with REINVENT 4.7 and newer. If for some reason it is not present, just copy synthsense folder into reinvent_plugins/components.

# Assuming REINVENT4 is installed
cp -r synthsense /path/to/REINVENT4/reinvent_plugins/components

Usage Instructions

  1. Copy these configuration files into your project directory and update them with absolute paths for your system:

    • examples/local_submitter.sh
    • examples/run.sh
    • examples/staged_learning
    • config/config_synthsense.yml
    • config/stocks_and_reactions_profiles.json

  2. Set your reaction and stock preferences in config_synthsense.yml.

  3. Submit the job:

    sbatch examples/run.sh

Note: the number of CPUs allocated in run.sh equals the nproc value defined in examples/local_submitter.sh.

Examples

See the examples/ directory for:

  • staged_learning.toml: Complete REINVENT staged learning configuration (all rewards inside)
  • run.sh: SLURM job submission script
  • reference_route_312_2110.json: Example reference route

Utilities

Generate Reference Routes

Use examples/generate_route.py to create reference route JSON files (with NameRxn classes) that can be used with RRScore:

python generate_route.py -r 3.1.2 2.1.10 -o 312_2110_two_step_route.json

Citation

If you use SynthSense in your research, please cite:

  • Dekleva, D.; Voronov, A.; Janet, J. P.; Ekborg, A.; Borišek, J.; Rambaher, M. H.; Loeffler, H. H. Synthesizability via Reward Engineering: Expanding Generative Molecular Design into Synthetic Space, ChemRxiv, 2025.

  • Loeffler, H. H.; He, J.; Tibo, A.; Janet, J. P.; Voronov, A.; Mervin, L. H.; Engkvist, O. REINVENT 4: Modern AI–Driven Generative Molecule Design. J. Cheminf. 2024, 16 (1), 20.

  • Genheden, S.; Thakkar, A.; Chadimová, V.; Reymond, J.-L.; Engkvist, O.; Bjerrum, E. AiZynthFinder: A Fast, Robust and Flexible Open-Source Software for Retrosynthetic Planning. J. Cheminf. 2020, 12, 70.

  • NextMove Software. NameRxn (Expert System for Named Reaction Identification and Classification). 2022.