Online Inquiry

Template-based Computer-Aided Retrosynthetic Route Planning


Reverse synthesis describes the iterative process of reducing a complex target molecule to a simple precursor by breaking the bond, that is, searching for possible precursors by starting from the product. The reverse synthesis method can effectively solve the synthesis problem of complex molecules and promote the development of organic synthesis science. Retrosynthetic route planning can be considered a rule-based reasoning procedure. The possibilities for each transformation are generated based on collected reaction rules, and then potential reaction routes are recommended by various optimization algorithms. Currently commonly used retrosynthetic route planning strategies are based on template and template-free computer-aided retrosynthetic route planning. The template-based computer-aided retrosynthetic route planning method refers to generate one or more candidate precursors by matching reaction rules with target molecules in which the templates can be sorted by experts or automatically extracted from the reaction database. With the advancement of systems biology experimental technology and the continuous accumulation of experimental data, the emergence of a large number of biomedical data provides strong support for data-driven biosynthetic design. Alfa Chemistry supports various machine learning methods to design template-based computer-aided retrosynthetic route for our customers.

Our Methodologies

  • Deep neural network (DNN) model

The hit rate of the target reactions predicted in the reaction verification set using the DNN model is relative high. At Alfa Chemistry, a DNN model is constructed to train the collected reaction data and we use this model to automatically extract reaction rules.

  • Monte Carlo tree search

We apply Monte Carlo tree search to speculate feasible synthetic routes with higher accuracy. Our scientists combine Monte Carlo tree search and DNN strategies to search for the retrosynthetic routes of multiple drug-like molecules.

  • Combine strategy network and Monte Carlo tree search

At Alfa Chemistry, combining strategy network and Monte Carlo tree search method is available to train the reaction data, which can significantly accelerate the speed and accuracy of retrosynthetic route prediction.

  • Similarity-based approach

An automatic retrosynthetic route prediction method based on analogous known reactions is designed, and the target molecule is synthesized by reverse transcription through imitating the synthesis method in the database.

Predicting Retrosynthetic Reactions Using Self-Corrected Transformer Neural Networks.Figure 1. Predicting Retrosynthetic Reactions Using Self-Corrected Transformer Neural Networks. (Zheng, S.; et al. 2019)

Our Process

At Alfa Chemistry, the process of template-based computer-aided retrosynthetic route design can be described as follows:

1. Prepare the reactions in the data set by adding atomic maps of reactants and products

  • Chemical reaction data set: The data set used in the retrosynthetic route design comes from well-known institutions and organizations, such as Elsevier, Chemical Abstracts Service.
  • Representation of the chemical reaction data: The quality of the chemical reaction modeling will directly affect the completion of subsequent tasks.

2. Extract transformation rules from reactions with atom mapping to guide the template-based computer-aided retrosynthetic route design

  • Atomic mapping: Atomic mapping is usually used to separate reactants and reagents in a reaction.

3. Predict and sort the precursors repeatedly according to the indicators of the evaluation route

  • Evaluation indicator: Most of our deep learning models used in the retrosynthetic route only apply a single standard Top-k, which refers to the percentage of standard precursors recorded in the data set in the first k recommendations.

4. Single-step retrosynthetic route design

In the single-step retrosynthetic route design, we use the template-based method to compare the target molecule with the template set to select a suitable reaction process. The template refers to the substructure mode that changes during the chemical reaction. We convert the product into the reactant according to the template in the template library.

5. Multi-step retrosynthetic route design

In order to enhance the design of the complete route and meet the practical requirements of the high complexity of the target molecule, it is necessary to improve the performance of the multi-step retrosynthetic route model. Our multi-step retrosynthetic route includes a single-step retrosynthetic route module that predicts the immediate precursor and a search planning module that recursively applies the single-step module. Moreover, Alfa Chemistry supports the Monte Carlo Tree Search (MCTS) method and other methods of tree search.

6. Template-based computer-aided retrosynthetic route output

Features of Our Template-based Computer-Aided Retrosynthetic Route Planning

  • High-quality data set
  • A machine learning model training
  • Computing abilities of template-based computer-aided retrosynthetic route planning

Our template-based computer-aided retrosynthetic route planning services remarkably reduce the cost, promote further experiments, and accelerate the process of drug design for customers worldwide. Our personalized and all-around services will satisfy your innovative study demands. If you are interested in our services, please don't hesitate to contact us. We are glad to cooperate with you and witness your success!


  • Zheng, S.; et al. Predicting Retrosynthetic Reactions Using Self-Corrected Transformer Neural Networks. Journal of Chemical Information and Modeling. 2019, 60(1).

Alfa Chemistry

  • Tel:
  • Fax:
  • Email:
Copyright © 2023 Alfa Chemistry. All rights reserved.