We went to two top chemistry conferences with a cutting-edge poster!
In July Piotr and I had the pleasure to participate in two conferences in the UK:
- Dial-a-Molecule Annual Meeting 2019, in York
- Machine Learning and AI in (Bio)chemical engineering, in Cambridge
Both events tackled the newest research in chemistry, with the latter especially focused on employing Artificial Intelligence in solving chemical tasks. The conferences were filled with interesting talks and workshops by researchers from leading academic institutions and life science companies. We would like to summarize some of the topics brought up during the talks.
Much promising research is being done towards simplifying and organizing the synthesis process. Prof. Martin Burke (University of Illinois) talked about “Democratizing Chemistry” by making it possible to synthesize vast amounts of organic compounds using only several basic building blocks. Jana Marie Weber (University of Cambridge) presented a few ways of identifying possible “Strategic Molecules” by representing known compounds and reactions as a graph and ranking the importance of its nodes.
There were interesting talks focused directly on solving chemical tasks with specific Machine Learning methods, such as Gaussian Processes, Bayesian Optimization or Reinforcement Learning. Prof. Alexei Lapkin (University of Cambridge, who was also the main organizer of the ML and AI conference) presented ways to combine statistics with advanced robotics to optimize the discovery of complex functional products, by finding the most promising experiments to perform and automating the synthesis process.
Some of the speakers took a broader look at the pros and cons of using AI in comparison to rule-based systems developed by experts. Joao Moreira (PSE) argued that mechanistic modeling and AI are mutually complementary and provided examples of the benefits of using both approaches. Matthew Robinson (University of Cambridge) took notice of the issue of uncertainty when it comes to measuring the performance of different methods, as well as the problem of scarcity of good benchmarks.
From fascinating talks and inspiring coffee-break discussions, we were able to gain useful insight into current challenges and possible directions in automating chemical synthesis planning. The events were attended by researchers with various scientific and professional backgrounds, who provided us with different views on the subject.
We had an opportunity to present our work on poster sessions during both of the events (we describe our poster in detail in another post). Our solution was met with considerable interest from synthetic chemists, as well as data engineers.
The team of Molecule.one at the Cambridge conference.