Retrosynthesis is a technique for solving problems in the planning of organic syntheses. This is achieved by transforming a target molecule into simpler precursor structures regardless of any potential reactivity/interaction with reagents, which is one of the most complex issues in the field of organic chemistry. With a huge number of available reactions accumulated, such as around 1.8 million reactions available from the United States Patent and Trademark Office (USPTO), how to find the most appropriate reactions for the target molecule with consideration of critical functional groups surrounding the reaction center, is one of main challenges in this field. To address this problem, one possible solution is to transform and represent such large amount of reaction data in a more meaningful way that is suitable for retrosynthesis design computationally. Inspired by the three-layer reaction model designed by InfoChem for their purpose of reaction classification, we propose to develop a multi-layer reaction knowledgebase to capture fundamental information for retrosynthesis. More specifically, we will define a reaction data model composing of three layers, namely the layers of reaction center (RC), maximum common structure (MCS) and specific reaction (SR), to capture full spectrum of reaction information. In addition, we will not only consider the backbone structure of RC and MCS, the rest of structures will also be represented as R-groups to allow R-group similarity comparison while reaction matching with the target molecule. Subsequently we will be able to develop the multi-layer reaction knowledgebase with accommodation of USPTO reaction data and from where to build up retrosynthesis applications.

Multi-layer reaction knowledgebase development towards retrosynthesis design

ACS Fall 2021

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