NNAASynth : Non-Natural Amino Acid Synthesizability Analysis

This repository provides an end-to-end framework for analyzing the synthetic accessibility of non-natural amino acids (NNAAs) and their protected forms. It leverages cheminformatics tools and deep learning models to:

• Generate protected forms of input amino acids
• Perform retrosynthetic analysis using AiZynthFinder
• Score synthetic routes using Chemformer and expert-augmented feasibility models
• Select the best synthetic route for each protected form

The workflow is designed to support peptide and amino acid design projects exploring beyond the standard 20 natural amino acids.

Features

• Automated protection of amino acids using customizable protection group rules
• Retrosynthetic analysis via AiZynthFinder
• Feasibility scoring using Chemformer and expert-augmented models
• Extensible and modular end-to-end pipeline for synthesizability analysis

Required External Repositories and Model Files

Before using this repository, you must manually install the following external tools and libraries from their respective GitHub repositories:

• AiZynthFinder
  Retrosynthetic planning software and its environment to run this project. GitHub: https://github.com/MolecularAI/aizynthfinder

• AiZynthModels
  Provides models and environment for AiZynthFinder and Chemformer API.
  GitHub: https://github.com/MolecularAI/aizynthmodels

• Chemformer
  Transformer-based model for reaction and molecule prediction. You will need to obtain only the vocabulary to run Chemformer API (bart_vocab_downstream.json) GitHub: https://github.com/MolecularAI/Chemformer

• Expert-Augmented Scorer Model Files To use the expert-augmented scoring models, download the following files into a folder named expert_augmented_models:

  mkdir expert_augmented_models
  cd expert_augmented_models
  wget https://zenodo.org/records/14533779/files/deepset_route_scoring_sdf.onnx?download=1 -O deepset_route_scoring_sdf.onnx
  wget https://zenodo.org/records/14533779/files/reaction_class_ranks.csv?download=1 -O reaction_class_ranks.csv
  wget https://zenodo.org/records/14533779/files/scscore_model_1024_bits.onnx?download=1 -O scscore_model_1024_bits.onnx
  wget https://raw.githubusercontent.com/MolecularAI/reaction_utils/refs/heads/route-scoring-example/examples/route-scoring/example-routes.json

Installation

1. Clone this repository

2. Create the environment from AiZynthFinder and AiZynthModels:

   • The aizynthfinder repository provides the environment to run this pipeline.
   • The aizynthmodels repository provides the environment to run the Chemformer API.

3. Install reaction-utils dependencies to both environments:

  cd ../reaction_utils
  poetry install --all-extras

4. Prepare the following resources (also described in the example.ipynb):

   • AiZynthFinder configuration file (example in nnaasynth/aizynthfinder_setup/aizynth_config.yml)
     • Download the public model and stock files as described in the AiZynthFinder repository.
   • Protection group and reaction rules CSVs (stored in reaction-utils/examples/nnaa/)
   • DeepSet and SCScore model files (described above)

5. Run the Chemformer API Chemformer API can be run locally with the bash script and the model checkpoint provided in the nnaasynth/chemformer_api folder. To run the API, you need to:

• set the required paths in chemformer_api.sh
• run the api and obtain the port number

   conda activate aizynthmodels
   bash chemformer_api.sh

• In case of a different model, please ensure it is compatible with the Chemformer API.

Project Structure

• example.ipynb — Jupyter notebook demonstrating usage
• run_synthesizability_analysis.py — Main pipeline class
• nnaasynth/ — Main package directory
  • aizynthfinder_setup/ — AiZynthFinder configuration and stock files
  • chemformer_api/ — Chemformer API-related code (if any)
  • dtos/ — Data transfer objects for amino acids and results
  • utils/ — Utility functions
  • workflow_components/ — Core modules for protection, scoring, and running retrosynthesis

Citation

If you use this codebase in your research, please cite the relevant tools (AiZynthFinder, Chemformer, reaction-utils) and our preprint:

@misc{geylan2025nnaasynth,
      title={From Concept to Chemistry: Integrating protection group strategy and reaction feasibility into non-natural amino acid synthesis planning}, 
      author={Gökçe Geylan and Mikhail Kabeshov and Samuel Genheden and Christos Kannas and Thierry Kogej and Leonardo De Maria and Florian David and Ola Engkvist},
      year={2025},
      eprint={10.26434/chemrxiv-2025-jtwn1},
      archivePrefix={ChemRxiv},
      url={https://chemrxiv.org/engage/chemrxiv/article-details/68592f151a8f9bdab5199312}, 
}

License

This project is licensed under the Apache 2.0 License. See the LICENSE file for more details.

Contact

For questions or contributions, please contact Gökçe Geylan at gokcegeylan96@gmail.com