Finding creative ways of synthesizing molecules is a long process. Chemists have to search for reproducible information through vast amounts of literature, look up papers and put pieces together one step at a time.

We wanted to offer a system that complements the chemists' knowledge and helps them consider alternative approaches. At the same time, we wanted to inspire the confidence required to determine what steps could produce the best possible drug.

We launched a web application that uses reactions from chemical databases to predict end-to-end synthetic routes. The chemists can explore the routes to unlock ideas and design more innovative while still reliable routes faster.

We designed for two user groups: discovery chemists and process chemists. Both are involved in the process of developing drugs but have different goals. While discovery chemists are interested in making small quantities that are pure enough for testing, process chemists try to optimise the results finding cheaper, safer and more scalable routes.

The project started with a stakeholder workshop and subject matter experts interviews. Together we defined the scope of the project and identified user pain-points. The first milestone was to create a proof-of-concept and understand the reactions of the researchers to the proposed solution.

I did secondary research to learn more about the current practices and the questions the chemists ask themselves during the process.

Even though we were a distributed team, we worked closely together. We conducted several workshops to define an end to end solution that would fit within the chemists workflow and we followed up with remote design reviews and feature definition sessions. We continued running user research sessions throughout the process.

A synthetic tree in ChemPlanner is made of known and predicted reactions. We played with iconography and colour to help users scan which reactions are predicted and which are found in the literature.

Predicted reactions take longer to generate than literature reactions; in order to provide users with results earlier we decided to display first syntheses made of known reactions while predictions were generated in the background.

ChemPlanner selects the most optimal solution based on overall yield and price but users can modify the routes by browsing alternative reactions. For each transformation there could be a large number of alternatives, so we developed a grid view to allow users to scan through the different options.

To further assist on this process we planned a feature to narrow down the number of options by flagging the ones that look more feasible. This would allow them to compare how various alternatives change the overall solution. However, due to project priorities this feature was not developed.

As we learned from user research, the reasoning behind the results is extremely important to help users assess the potential feasibility of the routes. The chemists have quick access to the data used to generate each transformation. These examples also give information of procedures previously used in similar transformations.

This was the first time I worked on a product all the way from kick-off to release and I was lucky to have the opportunity to work on an exciting project alongside very experienced people. We were a new and distributed team and we faced many challenges but were proactive and learned to overcome them. Eventually we worked out a very clearly defined process.

The most important takeaway I got from this project was learning to deal with risk, both from a product perspective and design point of view. Setting goals, understanding stakeholder motivations, not relying on a single design methodology but on the most appropriate to achieve the project goals, doing early technical analysis... If I look back I would have approached some things differently, but I am happy to keep the extremely valuable insights that I learned.