Reviews: Predicting Organic Reaction Outcomes with Weisfeiler-Lehman Network

Summary: This work provides a novel approach to predict the outcome of organic chemical reactions. A reaction can be computationally regarded as graph-prediction problem: given the input of several connected graphs (molecules), the model aims to predict a fully-connected graph (reaction product) that can be obtained by performing several graph edits (reaction) on some edges and nodes (reaction center) in the input graphs. Past reaction predictions involving exhaustively enumeration of reaction centers and fitting them to a large number of existing reaction templates, which is very inefficient and hard to scale. In this work, the author proposed a template-free method to predict the outcome. It is a 3 step pipeline: 1) identify the reaction center given the input graphs using a Weisfeiler-Lehman Network.