Retrieving compounds that satisfy desired chemical properties is a so-called inverse-QSAR problem. Recently, several approaches for the direct generation of structures of organic compounds by deep recurrent neural networks (LSTM, GRU) have been proposed. However, these approaches were initially designed for solving natural language processing problems, and have limitations applied to the generation of chemical structures. The general approach is the use of SMILES notation as a linear representation of organic structures, although it may lead to syntactically and chemically invalid compounds. Furthermore, there is a philosophical problem: <i>organic chemists think more in terms of functional groups rather than atoms and bonds.</i> To address these issues, we propose a molecular grammar approach: a molecule is constructed by sequential application of molecular grammar rules, whereby a grammar rule can be regarded as a structural fragment with “sticky bonds” and some additional criteria which guarantee connectivity between rules. In this approach, a generated molecule is structurally correct by design. We define a binary operation (+) for the concatenation of two rules into a larger one so that one can produce complicated building blocks from simple ones. We experimentally demonstrate that one can construct a sub-optimal grammar for a particular task. We investigated the applicability of molecular grammars to several inverse-QSAR problems and showed that this approach could effectively generate compounds with desired properties; however, the main advantage of the method is its universality.

Molecular grammars: Accurate and universal way to construct molecules <i>in silico</i>

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.

Molecular grammars: Accurate and universal way to construct molecules in silico