Achieving human-expert performance in predicting the outcomes of chemical reactions is a major open challenge in AI and chemistry. A solution to this challenge would have significant practical applications in areas ranging from drug design to atmospheric chemistry. However, in order to address this challenge, many issues need to be overcome including the lack of open data, the combinatorial and physical complexity of chemical reactions, and the need for interpretable solutions that illuminate the underlying reaction mechanisms. We will describe three projects aimed at addressing these challenges including the development and deployment of public databases of chemical reaction steps, and the development and training of deep graph neural network and transformer architectures to predict reaction outcomes in interpretable ways.