Benchmarking single step retrosynthesis models in a multi-step setting

Package to interchangeably use single-step retrosynthesis models as shown in Models Matter: The Impact of Single-Step Models on Synthesis Planning [1] as an extension to AiZynthFinder[2].

Overview

The SingleStepModelZoo allows the fast extension of AiZynthFinder[2] to interchageably use alternative single-step retrosynthesis models. The package includes Chemformer[3], LocalRetro[4] and MHNreact[5] as shown in [1], however additional single-step models can easily be incorporated. To use the models within the multi-step setting please refer to ModelsMatter.

Due to minor changes within the repositries, we fork the mentioned models which are included as submodules within the repository.

Installation

1. git clone --recurse-submodules https://github.com/PTorrenPeraire/modelsmatter_modelzoo
2. cd modelsmatter_modelzoo
3. a) If installing from scratch conda env create -f ssbenchmark.yaml
   b) If installing into AiZynthFinder instance conda env update -n [AiZynthFinder Environment Name] --file ssbenchmark.yaml
4. conda activate ssbenchmark or conda activate [AiZynthFinder Environment Name]
5. poetry install

By default, the environment includes all packages required for the use of the default models.

Using ModelZoo

Running default models

from ssbenchmark.model_zoo import ModelZoo
model = ModelZoo(single_step_model, single_step_module_path, single_step_use_gpu, single_step_settings)

The relvant variables are as follows:

• single_step_model: Each model is selected using their respetive (string) identifer, i.e. 'chemformer', 'mhnreact', 'localretro'.
• single_step_module_path: Path to the local installation of the single-step model package. If the selected model is prepared as a package this can be ignored, if the path is sent within the single_step_model class then this is not necessary.
• single_step_use_gpu: Boolean whether to carry out single-step model inference on gpu
• single_step_settings: Additional settings passed to the single-step model

Inference is carried out using the setup model using the standard model_call function. This requires a list of one or more smiles and returns the predicted reactants along with their respective probabilities as predicted by the single-step model.

reactants, probabilities = model.model_call(smiles)

All model classes can be found in ssbenchmark/ssmodels/.

Introducing new models

New models follow the same standard template, as shown in ssbenchmark/ssmodels/model_example.py. Importantly, each model class must include a model_setup function and a model_call function. Any string identifier can be used to identify the model, which is shown just before the start of the class and set as example_model in the example. The models can then be called using the same format as the default models, simply introducing the relevant model identifier and variables.

Using single-step models in multi-step framework

All models, both standard and any newly introduced models, can be easily used within the AiZynthFinder framework. Please reference the ModelsMatter or AiZynthFinder repositories for further details on the config files necessary. Public datasets and their respective trained models, as used in Models Matter, are availble on the Models Matter Figshare. For new models, the contents of the config file will be informed by the variables that are set in the ModelZoo instance.

License

The software is licensed under the MIT license.

Acknowledgments

This study was partially funded by the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie Innovative Training Network European Industrial Doctorate grant agreement No. 956832 “Advanced machine learning for Innovative Drug Discovery”.

References

1. P. Torren-Peraire, A. K. Hassen, S. Genheden, J. Verhoeven, D. Clevert, M. Preuss, and I. Tetko, “Models Matter: The Impact of Single-Step Retrosynthesis on Synthesis Planning,” arxiv, 2023. [Online]. Available: https://arxiv.org/abs/2308.05522
2. S. Genheden, A. Thakkar, V. Chadimová, J.-L. Reymond, O. Engkvist, and E. Bjerrum, “AiZynthFinder: a fast, robust and flexible open-source software for retrosynthetic planning,” Journal of Cheminformatics, vol. 12, no. 1, p. 70, 2020. [Online]. Available: https://doi.org/10.1186/s13321-020-00472-1
3. R. Irwin, S. Dimitriadis, J. He, and E. J. Bjerrum, “Chemformer: a pre-trained transformer for computational chemistry,” Machine Learning: Science and Technology, vol. 3, no. 1, p. 015022, 2022, publisher: IOP Publishing. [Online]. Available: https://doi.org/10.1088/2632-2153/ac3ffb
4. S. Chen and Y. Jung, “Deep Retrosynthetic Reaction Prediction using Local Reactivity and Global Attention,” JACS Au, vol. 1, no. 10, pp. 1612–1620, 2021, publisher: American Chemical Society. [Online]. Available: https://doi.org/10.1021/jacsau.1c00246
5. P. Seidl, P. Renz, N. Dyubankova, P. Neves, J. Verhoeven, J. K. Wegner, M. Segler, S. Hochreiter, and G. Klambauer, “Improving Few- and Zero-Shot Reaction Template Prediction Using Modern Hopfield Networks,” Journal of Chemical Information and Modeling, vol. 62, no. 9, pp. 2111–2120, 2022, publisher: American Chemical Society. [Online]. Available: https://doi.org/10.1021/acs.jcim.1c01065