Nowadays, computational planning of organic synthesis is of great interest, primarily due to the progress in deep learning. However, current approaches that are based on the generation of reactants and products as SMILES strings have some limitations: SMILES can be syntactically invalid, or a molecule itself can be invalid in terms of valency rules or broken cycles. To overcome these limitations, we used “molecular grammars” - templates for building organic compounds as a sequential application of grammar rules. One can regard these rules as molecular fragments with “sticky bonds” by which two rules can be matched together to construct a larger molecular fragment. These rules can be extracted from any chemical dataset by a deterministic algorithm. We used Transformer to generate either possible products of a reaction if reactants are known (direct mode) or for the generation of possible reactant structures for a compound of interest (reverse mode). The main advantages of our approach are (i) the model extracts all information from data; there is no need for explicit rules (ii) the model can operate with an arbitrary number of reagents and products (iii) the model is robust to noisy training data (iv) generated molecules are valid in terms of valency, cycles, and bonds. We compared the effectiveness of our approach with the IBM RXN model on the reaction dataset obtained from patents (USPTO). Our molecular grammar model outperformed the baseline IBM RXN model and achieved accuracy of top-1 direct prediction of 90% (vs. 89% for IBM RXN). In reverse mode, our model gained 77% for the top 5 predictions. We believe that molecular Transformer with molecular grammars can be an efficient tool for retrosynthetic analysis.

Retrosynthesis with transformer and molecular grammars

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.