AiZynthFinder 4.0: developments based on learnings from 3 years of industrial application

Abstract

We present an updated overview of the AiZynthFinder package for retrosynthesis planning. Since the first version was released in 2020, we have added a substantial number of new features based on user feedback. Feature enhancements include policies for filter reactions, support for any one-step retrosynthesis model, a scoring framework and several additional search algorithms. To exemplify the typical use-cases of the software and highlight some learnings, we perform a large-scale analysis on several hundred thousand target molecules from diverse sources. This analysis looks at for instance route shape, stock usage and exploitation of reaction space, and points out strengths and weaknesses of our retrosynthesis approach. The software is released as open-source for educational purposes as well as to provide a reference implementation of the core algorithms for synthesis prediction. We hope that releasing the software as open-source will further facilitate innovation in developing novel methods for synthetic route prediction. AiZynthFinder is a fast, robust and extensible open-source software and can be downloaded from https://github.com/MolecularAI/aizynthfinder .

Keywords: Computer-aided synthesis planning; Multi-step retrosynthesis; Open-source; Retrosynthesis software.

Fig. 1

The AiZynthFinder python package structure, outlining top-level modules and sub-packages

Fig. 2

Jupyter GUI for AiZynthFinder highlighting the route clustering. The relationship of the 20 routes extracted from the search of the Amenamevir drug is shown in a dendrogram. The bottom-part of the GUI shows a tab for each of the five clusters obtained when optimizing for the number of clusters. Each tab shows a pictorial representation of the routes

Fig. 3

The distribution of different reaction classes in the synthesis routes predicted for the different target sets

Fig. 4

The percentage of starting materials found in external or internal stocks for the AZ designs and Reinvent target sets, or ZINC and E-Molecules stocks for the ChEMBL and GDB target sets

Fig. 5

The number of unique templates used in the routes for different target sets. For AZ designs and Reinvent the model trained on Reaxys, Pistachio and AstraZeneca ELN is used and for ChEMBL and GDB the USPTO-based model is used (see Table 1)

Fig. 6

The distribution of the number of reaction examples per template for A the templates used in the routes for the AZ designs, B the templates used in the routes for the Reinvent targets, and C all templates in the internal AZ model. The y-axis indicates the percentage of the total number of templates